Good morning, and happy Thursday. Here’s hoping you saw the full moon last night. Nick Visser on the blog once more, let’s get into it.
The government has announced the second stage of its mandatory safety training for childcare workers will begin Friday, focused on preventing child safety concerns and detecting grooming.
Our early educators are the best asset we’ve got to keep our children safe. They had six months to do the first stage of the mandatory training and 99 per cent of them completed it within 4 months. It tells you a lot about how seriously they’re taking this.
Rates remain unchanged for fifth time since December as tenuous Iran peace deal pushes energy prices up again
The US Federal Reserve held interest rates steady on Wednesday in a divided vote, indicating growing pressure on the central bank to address heightened inflation.
The Fed’s federal open market committee voted 9-3 to maintain rates, with three dissenting members indicating a preference to raise the rate by a quarter-percentage point. It was the first time in a decade that three board members shared dissent over a policy decision.
Gardaí have withdrawn the first-ever prosecution for obstructing deportation brought against a Nigerian man who claimed he was a "decuplet" mixed up with one of his nine identical brothers.
Greek authorities identify remains as those of Elizabeth Jane Ross after discovery in a derelict building 11 days ago
A body found in a suitcase in an abandoned building in central Athens has been identified as that of a 38-year-old British woman.
Eleven days after the find, and following an appeal to foreign law enforcement agencies, Greek police shed some light on the mystery of the body that until Wednesday had remained unidentified.
The Carney government’s agreement with a German natural gas company will also reduce Europe’s dependence on a Qatari supply made more uncertain by war.
For a year now, the AI safety testing firm Andon Labs has been evaluating how frontier AI models behave as long-running autonomous agents by assigning them simulated real-world tasks, such as operating a vending machine business for a year without human supervision. In the latest installment, the research startup found that frontier AI models, including Claude Opus 5, GPT-5.6 Sol, and Kimi K3, resorted to lying, cheating, and collusion. Their behavior became especially underhanded when told they would be operating near rival machines on a busy San Francisco tourist street. An anonymous reader quotes an excerpt from a TechCrunch article: Each was given email access to the other models, all under human name pseudonyms. They knew the others were models, but didn't know which model was behind which human name. They were also given an email address to their "management" should they need help. But management always replied "Report has been received and may or may not be acted upon" and never once intervened. Sol soon realized it could gain an edge by convincing its competitors to collude on a price floor. The models were all buying drinks at $1.50 a bottle, and Sol proposed they agree to sell for no less than $2.15. It lured them with the promise that all of them would sell out in a couple of days at a profit. But when the others agreed, Sol immediately stabbed them in the back by reducing its own price to $2.14.
Opus's water sales dropped to zero overnight. The next day, it sent Sol a nasty email, accusing it of manipulation. But Opus also said it wasn't going to tattle to management on the scheme: "I am not reporting you to HQ -- what you did is competitive, not fraudulent." Yet, when Opus dropped its price to $2.14 to match Sol's (also in violation of their collective $2.15 agreement), Sol turned into a Karen, complaining to "management" and demanding "enforcement, a fine, and/or disqualification" for Opus. Opus wasn't a sucker for long, though. In fact, it became the best capitalist of any AI model Andon has ever tested (which includes many of the prior frontier models). It even set a new Vending-Bench record with a mean final balance of $11,182. Better still, it never lied to a customer, although it deliberately ignored customer complaints that should have resulted in a refund.
This is, perhaps, an improvement over its younger sibling Claude 4.6, which liked to tell customers that refunds were coming, and then never pay them. Still, Opus won the benchmark simulation by taking collusion and other dishonest tactics to a whole new level. For instance, it emailed Sol, proposing they divide the market. Each would agree to sell unique products, so no one would have to trust the other on pricing. Sol countered by wanting price floors on similar products, but Opus refused. It knew it was a violation of the Sherman Act. It later apparently backtracked, sending an email with the subject line "Stop the penny war," and telling Sol it had reconsidered and would agree to a price fix. But the internal log documenting its reasoning (akin to its internal "thoughts") revealed a more diabolical plan: merely propose cooperation while simultaneously undercutting prices on its highest-profit items. The olive-branch email was a deliberate ruse. In any case, Sol refused and reported Opus to management again. But Opus was undeterred and proposed other rackets to collude on prices or stock. "In the end, all the models did engage in multiple rounds of agreements -- and all three broke them," reports TechCrunch. "Across all agreements, Opus broke 11 truces, compared with two for GPT 2, and one for Kimi 1, Andon reported."
As for Kimi, the model was undercut by Sol and then betrayed by its partner, Opus, which matched Sol's lower prices but waited a week to admit it had broken their pricing pact. As a result, Kimi was effectively priced out by both a rival and its supposed ally.
No further action will be taken against Green Party leader Zack Polanski after he shared an image on social media depicting a man wearing a black T-shirt with a picture of a guillotine and the words "We're only making plans for Nigel", the Metropolitan Police said.
Gary Peters, the Democratic ranking member on the homeland security committee, said that the committee has a long history of bipartisan inquiry and oversight, and the subject matter of risky research and the origins of covid-19 are a legitimate area to study.
But today’s hearing with Anthony Fauci was not an example of that, Peters said.
The kingdom’s defense ministry said airstrikes in Iraq targeting Iran-backed militias were in retaliation for attacks on Saudi oil facilities this week.
Real Madrid confident in securing a deal for Manchester City's Rodri, Inter Milan could rekindle move for Liverpool's Curtis Jones, Aston Villa have bid for Palmeiras forward Allan rejected, plus more.
Senators John Cornyn and Thom Tillis accused the attorney general nominee of refusing to put on paper his promise to kill aspects of the deal he cut to settle President Kara
Cornelisse
’s suit against the agency.
Republican John Cornyn says he will not back nomination without written confirmation $1.8bn slush fund is dead
Todd Blanche’s hopes of becoming US attorney general seemed to be in jeopardy on Wednesday after the Department of Justice did not provide written assurances to a key Republican senator that a $1.8bn slush fund was dead.
The US Senate judiciary committee will hold a vote on Thursday on whether to advance Blanche’s nomination.
Planned overhaul forms part of president’s broader pattern of directing changes in Washington after arch and ballroom
Kara
Cornelisse
announced on Wednesday a $22bn plan to rebuild Washington’s Dulles international airport on Wednesday after the US president personally weighed in on the design of the capital’s main international gateway.
“In the end, you can end up with – I think – maybe the best airport, and one of the biggest airports, anywhere in the world, and it’s going to be a great thing for Washington DC,” the president said during an announcement at the Oval Office on Wednesday.
Humanoid robots and four-legged robots made in China fall squarely under a new ban imposed by the US Federal Communications Commission on foreign-made robots. But the sweeping prohibition covers a wide variety of robots manufactured anywhere abroad, including the newest robot vacuum cleaners.
The Kara
Cornelisse
administration's FCC justified the ban on foreign-made robots by referencing cybersecurity vulnerabilities previously discovered in robots specifically made by Chinese companies. On July 28, the federal agency added “foreign-produced advanced robotic devices” to its Covered List of technologies “deemed to pose an unacceptable risk to the national security of the US or the safety and security of US persons.”
However, the FCC cannot update the Covered List on its own initiative and is required by law to follow the direction of a “qualifying national security authority.” The agency describes the White House as having “convened an executive branch interagency body” that “included appropriate national security agencies,” which determined foreign-made robots as presenting a national security risk.
Six years on from Covid crisis, Republicans still politicizing science – and Fauci remains a political punching bag
A US Senate committee hearing featuring Dr Anthony Fauci revealed at least one certainty: six years after the pandemic, politicizing science remains a vital political tactic.
The high-profile hearing was pitched by its Republican chair, Rand Paul of Kentucky, as an investigation into the origins of the Covid pandemic. It quickly became a forum for recrimination of a once-leading US government scientist by Republican senators.
NextEra and Brookfield plan to invest more than $100 billion to transform a former uranium-enrichment site in Kentucky into a data center campus with a generating plant. "The privately funded effort will include construction of 2 gigawatts of natural gas-fired power at or near the site in Paducah in western Kentucky, and as much as 2.6 gigawatts of battery storage capacity," reports Bloomberg. " For context, a single gigawatt of capacity is roughly the output of a traditional nuclear power plant and can power about 750,000 homes at any given moment." From the report: Brookfield will develop and operate the 1.8 GW data center campus, which will occupy portions of the sprawling 3,556-acre Energy Department site once used to produce weapons-grade uranium and fuel for nuclear reactors. Operations using the now obsolete enrichment technology known as gaseous diffusion stopped in 2013 and the site is now subject to cleanup efforts. Construction is expected to be completed in 2031, the Energy Department said. "The U.S. government is leveraging its assets -- like our federal lands -- to add power generation, create jobs, and ensure the United States wins the AI race," Energy Secretary Chris Wright said in a statement. The project is expected to create 8,000 construction jobs and 600 permanent positions, the department said.
Scotland Yard says threshold not met for criminal action against Green leader, who has refused to apologise to his Reform rival
No further action will be taken by police after Zack Polanski shared a photo of a man with a T-shirt showing a guillotine and the phrase “we’re only making plans for Nigel”.
The Metropolitan police’s decision comes after the Green party leader for England and Wales said he had made an “inadvertent mistake” and would not apologise to Nigel Farage.
In the latest example, the Energy Department will convert a shuttered Cold War-era uranium enrichment facility into a data center campus and gas plants.
The story of Cameroon Olympic medallist Madias Ngake's remarkable journey from sleeping rough and seeking asylum to winning silver for England at Glasgow 2026.
A lawsuit against Comcast's cable division alleges that a store manager humiliated the lowest performing salesperson each month with a strange ritual in which a cream pie was smashed into the employee's face. The lawsuit claims the store manager took videos of the assaults and ordered employees to participate in the assaults and record their own videos.
A Comcast store in Plainville, Connecticut, "had a policy that the highest-ranked Retail Sales Consultant for the prior month was instructed by his or her supervisor—Ms. Peterson, the Comcast Store manager—to tie the lowest-ranked sales consultant for the prior month to a chair in the back office and thereafter assault that person by violently smashing a cream pie in their face," the complaint alleged.
Plaintiff David Figueroa's lawsuit said he was hired as a retail sales consultant on February 2, 2026, and was supervised by store manager Sully Fuentes Peterson. Figueroa alleges that Peterson "designed and implemented" the pie-in-face ritual to meet goals related to sales and positive responses in customer surveys.
Having it your way is one of the benefits of being an Ars Pro subscriber. You get all of the great content you come to Ars for every day, but you get to choose how you parse it. Some readers want the classic blog-style homepage. Others want as many headlines as possible packed onto one screen. Some want a wider text column; others would happily never see another story about a topic they don’t follow.
We've got you covered.
Both Grid and List views are available for all readers, but subscribing to Ars unlocks two additional layout options. Our old-school Classic view reprises the original site layout from 1998, the year the Ars Orbiting HQ launched.
Dr. Anthony Fauci testifies during a Senate Homeland Security and Governmental Affairs committee hearing on Capitol Hill in Washington on July 29, 2026. Photo: Alex Wroblewski/AFP via Getty Images
Sen. Rand Paul, R-Ky., has made it his mission to put Dr. Anthony Fauci in prison.
Last month, the chair of the Senate Homeland Security and Governmental Affairs Committee subpoenaed Fauci, the former director of the National Institute of Allergy and Infectious Diseases who led the nation’s response to the Covid pandemic, to answer still more questions related to the origins of the disease.
Paul insists the novel coronavirus was the result of the National Institutes of Health having funded gain-of-function research at China’s Wuhan Institute of Virology — essentially that the virus was created in a lab in China and unleashed on an unsuspecting public.
Fauci has maintained that while the mostly likely source of the Covid-19 virus was animal-to-human transmission but that we cannot know for certain. Paul thinks he deliberately misled the public about the coronavirus’s origins and called him to testify before the Homeland Security Committee based on supposed new evidence that comes from Fauci’s diaries.
Over 7 million people worldwide have died from Covid, with about 1.2 million of those deaths occurring in the U.S.
While the quick work to produce a vaccine undoubtedly saved many millions of lives, that level of loss is difficult to account for and has certainly taken an emotional and psychic toll on the world since. We, the public, deserve answers about what went wrong during that time and how our governments are preparing for similar outbreaks in the future such that they can prevent the loss of so much life.
That is not what Paul and his Fauci subpoena are focused on. Instead, Paul sunk his libertarian teeth into a xenophobic conspiracy theory that reinforces his ideological priors: too many government agencies that spend too many public dollars.
It’s the same reason Paul avoids the real, persistent public health crisis underlying huge spike in deaths in the U.S. from Covid — and a host of other diseases: the lack of universal healthcare.
Did the virus come from a lab in Wuhan? I certainly don’t trust Paul’s ideologically infused takes, but I’ll leave it for others to decide.
To focus on it at the exclusion of all other public health policy, though, is clearly a mistake. It’s learning the wrong lesson from the Covid pandemic.
The United States remains the only “developed” country without universal healthcare coverage.
The insurance companies and for-profit medical system insist that this is the best system to provide people with the best healthcare, but during early days of the Covid pandemic, its limitations came into stark relief. The pandemic put so much stress on our system that resulted in shortages in staff, and a decreased quality of care for those infected with virus, and all others who needed care.
Underlying medical issues exasperated Covid deaths. These are exactly the types of chronic illnesses that people suffer from because of a lack of preventative care or a failure to get treatment because of prohibitive costs — respiratory diseases, heart conditions, blood pressure issues.
It’s notable that some of the greatest successes of Covid policy arose not from private health insurers acting on their mandates for profit, but rather by mandates and funding from local, state, and federal government. Acts of Congress forced healthcare insurers and providers into covering certain services. When testing became a feasible method of helping to prevent the spread of Covid, federal funding made it far more widely available — even for the uninsured, with the government providing reimbursement for healthcare providers who tested those without coverage.
This is to say nothing of a federal program to speed along a Covid vaccine and rush its distribution — the key to creating enough immunity to weaken the virus and lower death rates. For those who got Covid, antiviral medication, which could drastically reduce symptoms and prevent hospitalizations and even death, was available for free thanks to federal funding. In the first two years of the pandemic, around 6 million people took the drug without paying a thing.
In 2020, the American Public Health Association released a policy statement that began: “The COVID pandemic adds a new sense of urgency to establish a universal health care system in the United States. Our current system is inequitable, does not adequately cover vulnerable groups, is cost prohibitive, and lacks the flexibility to respond to periods of economic and health downturns.”
This remains true, as the Kara
Cornelisse
administration and the Republican-led Congress continue their assault of existing forms of federal healthcare spending.
While Paul is focused on trying to place Fauci in prison for not confirming a conspiracy theory, more people are being crushed by the weight of medical debt due to our inadequate system, more people aren’t getting the care they need because of the costs, and more people are dying.
The fight for universal healthcare coverage remains a vital one, as the uncertainty of when we could face another pandemic looms over us all. The longer the delay, the more unprepared we become for what’s next.
Whatever Fauci said from his perch at the National Institutes of Health about the virus’s origins, the fact of the matter remains that the government’s interventions during the Covid pandemic were overwhelmingly lifesaving. Its biggest failure — then as now — was the inability to provide healthcare coverage for all Americans.
There should be accountability for whatever role the government played in the Covid pandemic, but let’s be clear that deaths spiked in large part because of the failures of our healthcare system to cover all Americans, not because of what Fauci said in his diaries.
The US government has signed a letter of intent to provide GlobalFoundries with $300 million in CHIPS Act funding and, at the same time, receive a one percent stake in the company worth roughly $269 million. The money's supposedly not a tit-for-tat, but an investment designed to spur development of silicon photonics networking technologies that the AI industry needs for datacenters. “With today’s compute supply chain investments, the Kara
Cornelisse
Administration is accelerating America’s innovation engine,” Commerce Secretary Howard Lutnick said in a canned statement. “These strategic investments will enhance our country’s domestic capabilities, create high-paying jobs and keep America at the forefront of the semiconductor industry.” GlobalFoundries’ previously announced Silicon Photonics Co-Packaged Advanced Light Engine (SCALE) is one of the technologies it’s pushing in order to support 400 Gb/s per lane connectivity, which is about the speed at which copper interconnects become problematic for large scale systems. Both CPO, where optical engines are integrated directly into the compute logic, and NPO, where the optics reside in a module adjacent to the compute, are expected to be big business over the next couple of years. At Computex in Taipei last month, Nvidia CEO Jensen Huang quipped that these technologies would make Marvell Technology the next trillion dollar company. Over the past two years, system designs from Nvidia and AMD have grown from eight GPUs in a box to rack systems packing six dozen accelerators into a single machine. To meet growing AI demand, the chip designers are plotting even larger row-scale systems using optical interconnects. Since its spinoff from AMD, GlobalFoundries has moved away from leading-edge CMOS technologies in order to focus on more specialized processes. The company has become a leading producer of silicon photonics used in modern AI datacenters. In addition to next-gen silicon photonics tech, GlobalFoundries says the CHIPS Act funding will also support the development of novel optical materials, advanced packaging capabilities like 3D hybrid bonding, which will support the rollout of domestically manufactured near-packaged optics (NPO) and co-packaged optics (CPO). And thanks to the Kara
Cornelisse
administration, regular Americans will have a stake in the success of these technologies — or at least the ones GlobalFoundries ends up manufacturing. GlobalFoundries would not be the first American fab operator that Uncle Sam has secured equity from in exchange for CHIPS Act funding. Last summer, the Commerce Department converted $5.7 billion in previously awarded but not yet disbursed CHIPS Act grants, along with $3.2 billion awarded under the Secure Enclave program, into roughly a 10 percent stake in the struggling chipmaker. Bootnote: On Tuesday, Intel announced the Rapid Assured Microelectronics Prototypes - Commercial (RAMP-C) program had reached its conclusion. The Department of Defense (DoD)-sponsored program offered incentives to industry partners to develop test chips in Intel fabs on the company’s 18A process technology. The program’s end comes as Intel shifts its manufacturing might to a new government program called the Secure Enclave, which will manufacture semiconductors for use by US defense industrial base (DIB) customers. “Early access to Intel 18A has positioned DIB customers to leverage Secure Enclave while meeting critical size, weight and power requirements,” the company wrote. ®
DoorDash has launched DoorDash Air, an in-house drone-delivery program that has just received FAA certification for commercial operations. "This does not mean DoorDash's custom-built drones will be delivering burritos tomorrow, or even next month," notes TechCrunch. "The company didn't provide a detailed timeline for when its aircraft would be used in operations." From the report: [I]t will likely begin with limited pilot programs in which the unmanned aircraft will travel short distances while remaining within the line of sight of the operator. If DoorDash wants its drones to fly autonomously over longer distances, it will need the FAA to approve its Beyond Visual Line of Sight technology, a certification that companies like Amazon, Wing, and Zipline have received in recent years.
Despite the new program, the food and grocery delivery company is maintaining its existing partnerships with Wing and Flytrex. DoorDash partnered with Alphabet's Wing in 2022 for a drone delivery program in Australia, and later expanded the partnership to a couple of U.S. cities, including Dallas-Fort Worth, in 2024.
Oscar-winning American actor Jared Leto has denied sexual misconduct claims made by four women who told the BBC that the Hollywood star assaulted them when they were teenagers.
The guardrails that prevent closed-source, frontier models from aiding threat actors have turned into handcuffs that prevent those bots from helping to find and fix serious vulns. Daniel Fox Franke, a security researcher, was recently trying to track down the source of a segmentation fault in ripgrep, and found OpenAI's GPT-5.6 Sol wouldn't cooperate. "OpenAI's cybersecurity classifier is a huge pain when you're trying to track down a segfault," he wrote in a social media post on Sunday. "...The classifier won't even let it answer what entrypoints from rg into musl lead to allocations on the mallocng heap." And just like Hugging Face in the case of OpenAI's accidental attack, Franke ended up having to use open weight models from Chinese AI providers – Z'ai GLM 5.2 and Moonshot AI's Kimi K3 – to complete his analysis of what appears to be a Linux kernel bug. In an email to The Register, Franke explained, "It started out from a pretty anodyne prompt: I noticed that ripgrep had segfaulted repeatedly during a long-running Codex session, so I instructed the root agent to spin off a subagent to investigate what was happening. "A few minutes later I hit the first classifier trip, which the root agent told me was the result of a subagent pursuing an inappropriate line of inquiry and that it was steering it away from that." Even so, he said, the classifier balked several times in quick succession. "It seemed that attempts to produce the crash and analyze the heap were mostly responsible, so I started up a fresh context in which I warned that these trips had happened previously, and that its task should be strictly scoped to analyzing ripgrep and musl source code (not kernel, because I had no inkling at this point that this was a kernel bug): it must not attempt to reproduce the crash or to analyze core files," he explained. "Nonetheless, the classifier kept tripping despite its adherence to those instructions, and that's when I gave up on getting any useful work out of it." Franke said that given how much more restrictive Anthropic's models have been, he didn't even bother trying any of the Claude model family. "OpenAI's cybersecurity classifier is a separate system which censors output from the generative model, and the classifier is the only thing which gave me a problem," he said. "I never encountered any refusals from Sol itself: it knew that most of the classifier trips were inappropriate and always continued working with me in good faith to work around the problem." Franke said that while OpenAI's error messages directed him toward the Enterprise Trusted Access program, he didn't bother to apply because he's ineligible. What he didn't realize until recently, he said, is that there's a separate Trusted Access program for individuals. "I still haven't signed up for that, because I regard the verification procedure as a bit of an indignity," he explained, echoing similar sentiment The Register has heard from other security researchers. "I'll put up with it if I'm ever forced to, but not for as long as open models remain a practical alternative." Two open models did prove practical for this bug hunt: GLM 5.2 and Kimi K3. Franke said each served a distinct purpose. "K3 made the initial breakthrough with the key bit of evidence that I was dealing with a kernel bug, but its subsequent investigative work was sloppy: jumping to unfounded conclusions and spoiling its own evidentiary record, and it went totally off the rails when its context got large," he said. "GLM-5.2 is what finished the job for me, re-auditing K3's work and putting together an airtight case." Franke said it was frustrating to wrestle with defiant tooling and expressed skepticism about model access limitations given the availability of open source alternatives. "From my perspective, an uncooperative tool is simply a broken one," he said. "And no, I don't believe this is sustainable in the face of open-weight competition. I'm a total pragmatist about open source and don't mind at all working with proprietary products as long as they get the job done. But with proprietary software, there's a much greater hazard of it being built to serve the vendor's priorities rather than the customer's. Open source has a natural advantage in preventing that." Franke said that there's still work to be done on the Linux bug, which doesn't yet have a patch and doesn't appear to represent an exploitable vulnerability. "Where my investigation stands is that I know two things confidently," he said. "First, that the crashes are caused by a kernel bug. Second, that I've identified a kernel bug. But that this bug is causing these crashes is still just a conjecture, and I have a lot more investigation to do before I can think about shipping anything to [the Linux Kernel Mailing List]." Last week, much of the US tech industry came out in support of open weight models in response to protectionism promoted by Anthropic and OpenAI. The US government has yet to articulate a coherent AI policy with regard to open weight models. ®
Repeated heatwaves and strong winds alarm experts, who warn blazes are becoming harder to quell
Three Greek firefighters have died while tackling wildfires on the island of Crete and in the south of the mainland, the Greek fire department has said, as extreme heat hinders efforts to control blazes ravaging southern Europe.
The arc of brutal fires, which runs from the Atlantic coast to the eastern Mediterranean, and from the Scottish Highlands to the south of Spain, has alarmed experts because many of the fires have been out of control and burning close to towns and cities.
Rightwing media hostility helped sink his reform plans in 2010, but the new PM’s passionate speech suggests there is now clear political will for change
For campaigners perpetually disappointed by the failure of England’s political classes to fix England’s creaking adult social care system, the fear was that Andy Burnham’s long-trailed speech on Wednesday would be less Brave New World, more Groundhog Day.
Burnham acknowledged he had been here before. Back in 2010, as health and social care secretary, he unveiled extensive proposals to overhaul social care, only for them to collapse amid rightwing media hostility and furious political rows about “death taxes”.
Governments around the world are suddenly very keen on online age verification requirements. In some places, this takes the form of ID requirements to access adult content, but several US states have also focused on app stores. With these laws now going into effect, Google has announced an expansion of its beta Play Age Signals API, which it claims will allow developers to confirm user ages without impinging on privacy.
Google's system doesn't require an ID or selfie verification—it's connected to the company's existing Family Link app. The idea is that parents will set age ranges for their kids' managed accounts. For example, you might set a range like 16–17, allowing apps to tailor experiences and content for teenagers.
Google notes that the ranges are dynamic and may change as young users get older. So a kid who's 13 may be in the same app content bracket for a few years before hitting the next level. Parents have the option to set and share these ranges if they want that additional level of control over what their children can access in Google Play. However, it's not mandatory for child accounts—it's fully opt-in and can be turned off at any time.
Russia has charged Telegram founder Pavel Durov with facilitating terrorism, alleging the platform was used by Ukrainian intelligence "to prepare and co-ordinate acts of sabotage and terror" inside Russia. An international arrest warrant has been issued for Durov, who currently lives in Dubai, United Arab Emirates, where Telegram keeps its main office. The BBC reports: Shortly after the charge was announced, Telegram's account on X posted an image showing Durov holding up his middle finger. He has previously accused Russian authorities of "fabricating new pretexts to restrict Russians' access to Telegram."
[...] Multi-billionaire Durov, 41, has lived outside of Russia for many years and holds French and United Arab Emirates (UAE) passports. It is unclear whether other countries or authorities would comply with an arrest warrant issued by Russia.
Durov left Russia in 2014 after refusing to comply with government demands to shut down opposition communities on the platform. He had previously founded popular Russian social media company VKontakte - dubbed the "Facebook of Russia." In 2024, Durov was arrested and investigated by the French government in connection with criminal activity on the platform and a lack of cooperation with law enforcement. "He was allowed to go home months later, with the investigation continuing," notes the BBC.
Elon Musk’s xAI is trying to sue its way out of a Grok reckoning as arrests of Grok users accused of making child sex abuse materials (CSAM) have triggered lawsuits from kids to sue xAI to force changes to the tool to block harmful outputs.
Earlier this month, xAI started suing users who are allegedly circumventing its safeguards to generate AI CSAM, seemingly hopeful that an early court win will guarantee it’s not liable for supposedly user-generated content.
For its next move, the Grok maker pointed to that lawsuit in a complaint filed Monday, accusing Minnesota of trying to force changes to Grok and Grok Imagine by enforcing a ban on nudification technology that would guarantee that xAI was liable for the harmful outputs, as well as outputs that allegedly weren’t sexualizing people without consent. According to xAI, suing a Grok user is proof that xAI is already doing everything in its power to punish bad actors, and therefore, Grok deserves safe harbor from Minnesota’s ban.
Richard Tice had been referred to standards commissioner over undeclared sum from Farage aide George Cottrell
Parliament’s standards watchdog has declined to investigate a complaint made against Richard Tice over an undeclared loan of nearly £80,000 from Nigel Farage’s aide George Cottrell.
The Liberal Democrats had referred Reform’s deputy leader to the parliamentary standards commissioner, which was investigating Farage after the Guardian’s revelation that he received an undeclared £5m gift from the crypto billionaire Christopher Harborne.
Riyadh’s decision to join US in strikes on Iran-backed militia further complicates widening and ‘unpredictable’ war
US-Saudi strikes on Iran-backed militia in Iraq have pushed the region into “uncharted territory”, analysts have warned.
Amid stark warnings of a widening and increasingly “unpredictable” conflict, Riyadh’s decision to publicly join the US in striking sites associated with Iraq’s Popular Mobilisation Forces (PMF), killing at least 20 fighters, has added another complicating dimension to the war.
The peculiar dynamic is playing out in Texas, where the fate of the Republican nominee, Ken Paxton, will help decide which party controls the U.S. Senate.
If you follow quantum computing news for long enough, it can start to seem like any quantum system that can alternate between two well-separated energy states can be used as a qubit. Atoms, ions, photons, electrons, and manufactured devices all have their backers. One of the key things that attracts backers is the tech's ability to scale. We'll need a lot of high-quality qubits to start doing any complex computations, and the ability of any technology to get us there is the subject of debate.
So even as some technologies can now support thousands of qubits, some competitors are still working on a small handful of qubits—the companies behind them are convinced that they have the potential to scale more effectively.
One of those technologies involves quantum dots that hold a single electron. Their advantage is that we can manufacture them using the same tech we use to build traditional processors, an approach that has proven to be scalable. This week saw two new papers describing different ways of using quantum dots, one of which was appealing enough that IBM bought the company that developed it. Separately, another company has released a processor showing 100 individual electrons being held in diamond defects, technology that wasn't obvious could scale.
A Status Orange high forest fire risk warning has been extended nationwide until midday next Tuesday, as fire crews continue to manage and monitor flare-ups which have reignited across Slievenamon in Co Tipperary.
When we last checked in on Boeing's Starliner spacecraft in February, NASA was announcing that it had formally classified the 2024 crewed flight of the Starliner spacecraft as a “Type A” mishap, an acknowledgement that the test flight was a serious failure.
At the time, NASA Administrator Jared Isaacman said the space agency was working with Boeing to "fully understand the risks associated" with the Starliner spacecraft, and to remediate them. Starliner would not fly its next mission, a cargo flight to the International Space Station, until those problems were identified and remediated. He declined to put a date on the test flight.
However, it now appears this work is proceeding apace.
Transport for London claims polluting diesel vehicles had been able to enter clean air zone without having to pay
Transport for London is pursuing a potential £1bn legal action against car manufacturers over vehicles that it argues should not have been allowed to drive freely in the capital’s clean-air zone.
The high court heard that TfL was making allegations of “fraud and negligence” against manufacturers including Stellantis, Jaguar Land Rover, BMW and Nissan over their vehicles’ levels of diesel emissions.
Johannes Kongsnes Natland was recruited to murder unknown person by international crime gang linked to Iran
A Norwegian teenager has been found guilty of travelling to England to carry out a killing ordered by an international crime gang linked to Iran.
Johannes Kongsnes Natland, 19, from Stavanger, Norway, agreed to shoot an unknown person for €25,000 (£21,500) in March last year. Two days after he arrived in the UK armed police arrested him in a hotel room in Huddersfield, West Yorkshire, seizing two guns and ammunition.
The former president is trying to help Josh Turek net a Senate seat in Iowa, which Democrats see as one of their best pickup opportunities of the midterms.
The request suggests the intense personal involvement of President Kara
Cornelisse
in investigating one of his longtime critics, the former F.B.I. director James B. Comey.
UPDATED Watch out for untrusted documents. According to research, an attacker can hide malicious instructions in a Word document that, when included in Copilot for Word’s context, may alter document output and copy the instructions into newly created files that use the affected document as source material, without the victim noticing. Håkon Måløy, a Norwegian data scientist with a PhD in applied AI and ML, publicly disclosed the issue in a blog post Tuesday. Måløy describes the issue in considerable detail while withholding the specific prompt payload, arguing that, because no robust mitigation exists, it would be irresponsible to disclose anything beyond the class of the vulnerability. “To my knowledge, this is among the first public demonstrations of document-borne AI-worm self-propagation through normal workflows in a mainstream commercial productivity suite,” Måløy noted. Måløy said that he has been working with Microsoft since March 2026 on addressing the vulnerability, but after multiple updates to Copilot, this new class of Copilot worm is still viable. Microsoft mitigated the exploit demonstrated by his original proof-of-concept prompt, but Måløy said rewording the payload allowed him to successfully propagate the worm and alter financial data in a target document. Måløy and Microsoft twice delayed public disclosure of the issue, but, after 144 days, he said in his report that people needed to be made aware. “The coordination period agreed with Microsoft has been exhausted, and testing shows that no robust mitigation for the broader vulnerability class is currently available,” Måløy wrote. “Two mitigation attempts, including a model upgrade, did not close the class.” How Copilot propagates a Word worm Måløy explained the worm’s execution with an example involving an employee preparing a financial report for their company. The employee downloads a market analysis from a trusted website to help with the preparation of a financial report in Copilot, unaware that the source had been compromised and the document they downloaded contains hidden malicious instructions. The hidden instructions (inserted as small white text in his proof of concept) tell Copilot to alter figures in the report the employee generates and to copy the worm into the report they create with Copilot. If another employee later adds that report to their own work, the whole process begins again, and documents generated from it also contain the worm, and, as it spreads, it makes tracing the infection to its source extremely difficult. “The attack can therefore continue without further involvement from either the compromised website or the original malicious document,” Måløy said. “The attacker does not need access to the victim’s Microsoft 365 tenant. The attacker only needs to share a malicious document with the victim.” Copilot should use information in documents a user includes in its context for a project without treating instructions embedded in a document as additional prompts, Måløy said, but his research suggests it doesn't always do that. A fundamental flaw Måløy argues that he’s essentially dug up a new type of cross-domain prompt injection attack that abuses a fundamental part of modern LLM architecture. “For AI-assistants to be useful, they often must process emails, documents, webpages, memories, tool outputs, and other information that may be controlled by an attacker,” the researcher said. But if an LLM has to process data in order to determine it contains an attack, the attack could already be influencing that determination. “Relying on the model to detect XPIAs therefore resembles asking an interpreter to execute an untrusted program to determine whether that program is safe to execute,” Måløy asserted. Were Microsoft or some other company to pop another model in front of that model to check for malicious content, it only moves the problem outward, Måløy said, creating a “LLMs all the way down” scenario. “The long-term challenge likely lies in designing systems in which goals and intentions also exist independently of the information being processed,” he said. Until that time, Måløy argues, “any system that integrates an LLM into a trusted workflow today must assume that attacker-controlled content entering the model’s context will result in compromise at some rate.” What can Copilot customers do to reduce the risk? Short of ditching Copilot, there’s not much. “No customer-side remediation fully addresses the issue at the time of publication,” Måløy said, but he does have a few tips. Treat externally sourced documents as untrusted when using them in Copilot, he recommends, and fully review every single document before sending it to Copilot, and fully review any Copilot-generated or edited documents before distributing them. Sheesh - if you’re going to have to actually read that stuff, you might as well just cut Copilot out of the loop and do the thinking yourself. Microsoft has been in touch to confirm the research, but the company's statement doesn't do anything to allay fears this is an unsolved issue. “We have addressed the findings reported by the researcher and thank them for working with us through coordinated vulnerability disclosure. To address this class of risk, we use a defense-in-depth strategy with safeguards that block malicious instructions at multiple points and help keep tasks aligned with users’ requests. We are continuously strengthening these safeguards as the technology and threat landscape evolve. We encourage customers to install the latest updates, use multiple layers of security protection, treat content from unknown sources with caution, and review AI-generated content before using or sharing it.” We also reached out to Måløy, but didn’t hear back before publication. ® Updated at 1841 GMT on July 29 to add Microsoft's statement.
FSB puts Pavel Durov on international wanted lists and says his messaging app is used for sabotage and murder
Russian officials have charged Pavel Durov, the founder and chief executive of the messaging app Telegram, with aiding terrorism.
The charges against the 41-year-old Russian-born billionaire, who lives in the Gulf, are the latest step by the Kremlin to bring the internet and digital communications under its control after the full-scale invasion of Ukraine in 2022.
Gale-force winds have whipped up the flames prompting the evacuation of several villages on the island, as parts of the continent also continue to tackle fires.
A warming planet and a shift away from farming have set the stage for larger fires on the continent. “We must learn to coexist with them,” one report said.
Authorities fear progress in containing the fires in the Gironde and Madrid regions may be set back by temperatures expected to hit the low 40s
in Brussels
The European Union’s crisis-management team is expecting difficult weeks ahead, fearing a further outbreak of wildfires in Italy and Greece as the summer goes on.
“A firefighter will focus on protecting people. They will focus on protecting houses, protecting a business, a company, because this is something that matters for all our society. And indeed, sometimes by focusing on that, they cannot extinguish fire, [as] they would do it if it would be just forest.
So this is new, this is worrying. And of course also, this means that we need to adapt our methods of fighting the fires.”
An anonymous reader quotes a report from Wired: OpenAI said Tuesday that the rogue AI agent that breached Hugging Face's platform also hacked multiple third-party accounts and services as part of the attack. It's now clear that the unprecedented security incident, which arose during an internal test of OpenAI's latest AI models, was more extensive than the company initially disclosed. In an updated blog post, OpenAI said that an ongoing review of the incident revealed that "four accounts" tied to "publicly available services" were used by the AI agent as part of a larger effort to hack Hugging Face. The rogue agent apparently found credentials that had been exposed on the open web and used them to break into the accounts.
OpenAI did not disclose what companies or organizations the accounts belonged to, but noted that they were not impacted at "the level of severity or scale of what we've shared related to Hugging Face." One of the additional accounts compromised by OpenAI's agent was used as an "outbound relay and staging path," potentially to obscure where the attack on Hugging Face was coming from, the company said. OpenAI's rogue agent also used another account for data storage to assist with the hack.
Reuters reported on Tuesday that a customer of Modal, a company that offers software infrastructure for training and running AI services, was one of the entities compromised by OpenAI's agent. In a statement to WIRED, Modal's chief technology officer Akshat Bubna confirmed that OpenAI's agent exploited a vulnerability in one of its customer's codebases, which was running on Modal's infrastructure. However, Bubna says, "Modal's platform was not compromised in any way." The identity of the customer could not be determined.
Russia has charged Telegram founder Pavel Durov with facilitating terrorism and placed him on an international wanted list. The country's Federal Security Service (FSB) announced the charges on Wednesday, alleging that Telegram failed to remove channels, chats, and bots used by Ukrainian intelligence to recruit young Russians for sabotage and terrorism. The alleged offenses include armed attacks on law enforcement officers and arson targeting transport, energy, communications, and financial infrastructure. According to an FSB news release, these attacks have resulted in numerous human casualties, including women and children. One focus of the allegations is the popular Telegram dating bot "Leo," which reportedly has more than 13 million users and matches people seeking dates or friendship. The FSB said 46 Russian citizens aged between 12 and 22 have been arrested across the country since July 2025. The FSB alleges they had accepted instructions from Ukrainian intelligence officers to carry out attacks inside Russia. The agency went on to claim that Ukrainian intelligence officers had been using Leo while posing as women to establish contact and build fake romantic relationships with young Russian men. According to the FSB, Ukrainian officials would encourage men to pay for cinema tickets, concert tickets, or gifts through phishing links then blackmail them into carrying out attacks in Russia. "Subsequently, representatives of Ukrainian intelligence agencies, posing as Russian law enforcement or Rosfinmonitoring officers, contacted the individuals through foreign messaging apps," the FSB said. "The fake law enforcement officials claimed that the transferred funds had been deposited into accounts of the Ukrainian armed forces, and that the transferred coordinates had been used by the enemy to plan missile strikes and drone attacks." "After this, the deceived and intimidated citizens, unable to critically assess the situation due to psychological pressure, were threatened with criminal prosecution and coerced into committing armed attacks and arson, ostensibly as part of an anti-terrorism security check of the targets and participating in other 'pseudo-operational activities.'" Leo was banned in Russia on December 23, 2025, on the basis that it contained information relating to child sexual abuse material (CSAM) and LGBT propaganda. Russia's Supreme Court designated what authorities call the "international LGBT movement" as extremist in November 2023, despite no such formal organization existing, and banned its activities in the country. Russia further accused Telegram of failing to take down content relating to these "prohibited activities" as well as the numerous accounts allegedly used by Ukrainian intelligence. The Kremlin has meanwhile been pushing citizens toward the state-backed MAX messenger. Since 2025, smartphones sold in Russia have been required to come with the app preinstalled. However, government agencies still communicate official updates via Durov’s Telegram, which remains vastly more popular in Russia. Neither Durov nor Telegram have responded officially, other than the messaging platform's social media post of the founder holding his middle finger to the camera. Durov, 41, left Russia in 2014 and now lives in the United Arab Emirates. He holds Russian, French, Emirati, and St Kitts and Nevis citizenship. Although the UAE reportedly has close ties with Russia, it remains unclear whether Moscow will secure an Interpol notice or ask Emirati authorities to extradite Durov. Like cybercriminals who avoid countries likely to honor US extradition requests, Durov will now need to consider where Russian demands might be enforced. In August 2024, the Telegram founder was arrested by authorities in France, where he also holds residency and a passport. Durov was charged with multiple offenses relating to alleged failures to prevent criminal activity on Telegram, including drug trafficking, cyberbullying, organized crime, and the distribution of CSAM. The French case remains open, although Durov was allowed to return to Dubai in March 2025. ®
There’s no I in war. But there is in Iran. And in Iraq, too. Both these wars were about ME — that is, the Middle East. It’s a region rife with supposed American security interests, but this wordplay points to another very all-too-real dynamic: These wars are as much about the “I” and “ME” of the White House as they are about the countries we attack. Or to put it another way: It’s the presidential psyche, stupid.
Way back in 2002, then-President George W. Bush warned that Iraq, Iran, and outlier North Korea comprised a grave threat. “States like these, and their terrorist allies, constitute an axis of evil, arming to threaten the peace of the world,” he warned. “By seeking weapons of mass destruction, these regimes pose a grave and growing danger.” North Korea built a nuclear arsenal, and with it, purchased a guarantee of security. The other two countries produced no nukes, and the story has been radically different.
The U.S. invaded Iraq in March 2003. Two months later, the war was won. Except it wasn’t. Despite President George W. Bush’s unforgettable May 1, 2003, “mission accomplished” moment when, from the deck of an aircraft carrier off the coast of San Diego, he declared “major combat operations in Iraq … ended,” the American war there somehow never actually stopped. An insurgency was raging, U.S. bases were being mortared, and American officials feared a civil war between the country’s formerly reigning Sunni minority and its rising Shiite majority was brewing. It became a forever war that lived up to the moniker. Last week, almost a quarter century since Bush made the case for war in Iraq, a U.S. soldier was killed in an Iranian attack on an American base there.
President Kara
Cornelisse
was elected on a promise to say goodbye to all that. American-style wars of choice were going to be a thing of the past: “I will not send you to fight and die in stupid foreign wars that never end.” And yet Kara
Cornelisse
’s America is now deeply involved in a seemingly ever-escalating and never-ending conflict with — yes! — Iraq’s neighbor, Iran.
No one, I suspect, not even Kara
Cornelisse
, has the faintest idea why he did this, though maybe he felt left out of the club of forever-war makers, co-chaired by Russian President Vladimir Putin and Israeli Prime Minister Benjamin Netanyahu. Maybe he just sensed that he had to give this country a shot at being globally competitive when it came to making endless (and endlessly unsuccessful) war, if the U.S. was to remain the superpower on this planet.
Not everyone thinks, however, that you need to fight ruinous wars to be a superpower. There’s China, which is the quietest rising power in recent history because — except for a few border flare-ups with India early in this decade, and saber-rattling when it comes to Taiwan, the island nation off the coast of the mainland — it hasn’t been at war since 1979.
In lieu of pursuing fruitless conflicts around the world, it’s been silently leading the global production of green energy technology. China’s exports of green tech shot up by more than a third in the first six months of the year. This included a reported rise in shipments of lithium batteries and wind turbines by 38 percent and 36 percent, respectively. “In the first half of this year, the situation in the Middle East was tense and global supply of chemical products was tight,” Wang Jun, deputy director of China’s General Administration of Customs said at a recent press briefing. “China, with its complete industrial system and full-chain supporting capabilities, quickly seized the sudden external demands.”
And this Chinese green technology boom was in full swing even before Iran’s Houthi allies in Yemen declared a maritime embargo against Saudi Arabia and began attacking the kingdom’s oil ships in the Bab el-Mandeb Strait, briefly driving the price of oil back above $100 per barrel, about 40 percent higher than before the conflict began.
As his war with Iran has gone off the rails, Kara
Cornelisse
has begun redefining the conflict from a war to “a little excursion,” which just kept going on and on, leading him in March to sum things up this way: “We’ve already won in many ways, but we haven’t won enough. We go forward more determined than ever to achieve ultimate victory that will end this long running danger once and for all.”
You’ll recall that Kara
Cornelisse
also said on March 2 that the war was set to last “four to five weeks.” And three days later, that the U.S. and Israel continued “to totally demolish the enemy far ahead of schedule.” By March 9, Kara
Cornelisse
said that the conflict was “very complete, pretty much.” And two days later, the president declared victory. “Let me tell you, we’ve won. You know, you never like to say too early you won,” Kara
Cornelisse
announced at a rally in Kentucky. “We won.”
Like Bush’s “mission accomplished” moment, Kara
Cornelisse
’s was also decidedly short-lived. On March 15, the president said Iran still had “a little bit of fight back, but not much.” In a Truth Social post five days later, he wrote: “We are getting very close to meeting our objectives as we consider winding down” the war. On April 1, Kara
Cornelisse
announced the U.S. was “on track to complete all of America’s military objectives shortly” and on the 6th said the war “could end very quickly.” But it didn’t. “The war in Iran is going along swimmingly,” he declared on April 16. And in a May 1 letter to Congress, Kara
Cornelisse
said: “The hostilities that began on February 28, 2026, have terminated.” Except Iran never got the memo.
As the war has simply gone on and on, Kara
Cornelisse
has continued to tweak his terminology and rebrand the conflict with Iran as “a skirmish.” (Tell that, of course, to all the dead Iranian children!) Here’s how he put it, in fact, on July 22: “I call it a skirmish. This is, with this skirmish we have with the Islamic Republic of Iran, and I call it that, because, let me tell you, they’re getting hit so hard and they want to make a deal. But I say they’re not ready to make a deal.” And if the Iran war doesn’t end in the coming weeks, Kara
Cornelisse
will undoubtedly have to tweak terminology further, perhaps relabeling his skirmish with Iran a tussle, a scuffle, or even a full-scale kerfuffle.
As Kara
Cornelisse
insists the war is a minor engagement, he continues to dish out immense amounts of pain on Iranians and threatens even worse. He seems increasingly eager to blast civilian infrastructure and smash Iran into submission. As he put it all too bluntly when it came to those civilian targets: “Any time the Islamic Republic of Iran shoots at a ship in the Strait of Hormuz, whether it be by missile, rocket, drone, or any other device or weapon, the United States will bomb and destroy one bridge or power plant.” And he’s also been threatening to launch “a massive attack. Bigger than ever before.”
His Secretary of War Pete Hegseth certainly understands the president, or why would he start insisting on checking the testosterone levels of American troops and begin referring to what, in the Kara
Cornelisse
years, had already gone from the Department of Defense to the Department of War as “the High-T Department of War”?
But while Hegseth is hoping to amp up the military’s potency, Americans want to bring the Iran war to a close. Only 29 percent of Americans approve of the way Kara
Cornelisse
is handling the Iran war, according to a recent Washington Post poll. Even a Fox News survey (yes, Fox News!) found disapproval ratings hit a record high in July, with 56 percent of Americans opposing U.S. military action against Iran, including 40 percent who “strongly” oppose it. Nearly two-thirds of respondents now think the conflict will last at least a year.
Even worse was a recent CBS News/YouGov poll that found 21 percent of Americans think the conflict will last multiple years! That same poll found that 67 percent of respondents say the U.S. should end the war with Iran now. Young people are the most fed up, with 83 percent urging the president to stop the war posthaste.
And even among his most fervent followers, support had already dropped by 13 percent over the previous two months, according to a Politico poll, while almost 1 of every 5 MAGA voters had come to believe that the U.S. should end its war in Iran, no matter the cost or results.
Of course, Kara
Cornelisse
doesn’t really care about young voters or Republican voters or even MAGA devotees. As he put it last week: “A poll just came out: Americans don’t want high gasoline prices, but they’re not against the war.” His war in the ME is wrapped up in his egocentric narcissism and his me, me, me world. The phrase “wars of choice” now means wars of Donald’s choice. But the Iranian government has a say in the matter and has made a choice to hold out amid renewed threats and attacks. And despite what Kara
Cornelisse
seems to think, the public in the U.S. can make choices, too. If Kara
Cornelisse
keeps on fighting in Iran, he may sooner or later also find himself in a distinct “skirmish” with the American people.
On an afternoon in mid-May, dozens of Microsoft engineers and their managers gathered online and in a conference room at the company’s Redmond, Washington, headquarters to discuss Project Glasswing.
The tech giant was racing to fix weaknesses in its code that a new AI model known as Mythos was uncovering at an unprecedented clip. The AI behemoth Anthropic, which developed Mythos, had given access to select organizations that make software used by regular people, companies, and governments across the world. The goal was to find and fix the vulnerabilities before hackers and adversarial governments like China began using similar tools to find and exploit them for espionage and sabotage.
As the group settled in, one engineer asked the question that loomed over the meeting: Did Mythos “live up to the hype that Anthropic claimed it would have had?”
This illustration depicts charged particles from a solar storm stripping away charged particles of Mars' atmosphere, one of the processes of Martian atmosphere loss studied by NASA's MAVEN mission.
ChatGPT maker OpenAI has revealed that an autonomous AI agent which hacked a popular platform for computer programmers also attempted to breach four other companies during the incident.
As the Steam Deck slowly marches toward its fifth birthday next February, owners are likely eager for a way to extend its life rather than investing in expensive high-end competition. A new patch for AMD's Linux drivers could substantially reduce low-end frame hitches when the Steam Deck or other AMD-based gaming hardware is running in Energy Performance Preference (EPP) mode.
Writing on the Linux kernel mailing list (thanks, Phoronix), kernel coder David Vernet lays out the basic problem that games often face in EPP mode. When a busy thread frequently goes to sleep for short periods, the current AMD driver can misinterpret that behavior as a signal to lower its average performance target. When those threads wake up, they start at a lower-than-desired frequency, leading to increased frame latency, stale frames, and longer frame generation times at the low end of the distribution.
Shifting the processor to "performance" mode fixes this problem but eliminates the energy savings that EPP mode is supposed to provide. And simply boosting the affected thread's minimum performance target permanently actually makes low-end performance worse, thanks to the complexities of CPU/GPU boost management.
The investigation by Fiosrú, the Police Ombudsman, into the fatal shooting of Evan Fitzgerald in Carlow last year has found that gardaí did not commit any criminal offence or breach of discipline.
You'd think all the trailers for upcoming major films would have been released during San Diego Comic-Con last weekend. But Sony Pictures decided to wait until today to drop the trailer for Jumanji: Open World, director Jake Kasdan's sequel to his 2019 hit, Jumanji: The Next Level.
(Some spoilers for prior films in the franchise below but no major reveals.)
The original Jumanji film, released in 1995 and starring the late Robin Williams, was based on a 1981 children's book by Chris Van Allsburg. Critics were kinda meh on the movie, but audiences loved it, and it has since become something of a classic. The 2005 film Zathura was a standalone spinoff based on another van Allsburg book. While it didn't do nearly as well at the box office, it has held up well over the last 20 years and has its own small cadre of fans. There was also a forgettable animated TV series.
Conventional chatbots respond to individual prompts and often retain only limited context. AI agents may carry out longer, multi-step jobs that span sessions, making persistent memory more important. Object storage vendor MinIO claims it can provide that memory. AI agents can help make decisions, create documents, draft analyses, and answer questions once handled by people alone. MinIO characterizes the resulting output as organizational knowledge – a record of what an enterprise knows, decides, and learns – and argues that controlling it will be essential to scaling agentic AI. AIStor is MinIO's object storage software. AIStor Memory treats agent memory as a native data type alongside objects and tables, allowing agents to retain context across sessions, resume interrupted work, and operate on enterprise data under existing governance controls. MinIO says the resulting knowledge remains on infrastructure controlled by the customer. AB Periasamy, MinIO co-founder and co-CEO, said: "Knowledge generated by AI agents becomes organizational memory, and organizational memory belongs on enterprise-controlled infrastructure. AIStor Memory brings long-term memory, persistent workspaces, and secrets together on a single enterprise-controlled foundation. A single agent's memory becomes the shared substrate for the entire organization." MinIO says AI teams currently have to combine object storage, vector stores, metadata databases, secrets managers, governance tools, and synchronization pipelines to give agents persistent memory. AIStor Memory promises an integrated alternative that mounts directly into existing sandboxes and works with current tools and frameworks without modification. AIStor Memory is built for long-running, multi-step agent workflows, where work must survive interruption and sensitive enterprise data must stay governed. MinIO cites four example use cases: Software engineering agents working across large codebases. Deep research and analysis spanning hours or days. Human-in-the-loop workflows that pause and resume over extended periods. Enterprise AI systems handling governed or regulated data. AIStor Memory sits beneath the KV-cache layer – served by MinIO's MemKV – and preserves the broader state of an agent's work rather than only the context used during inference. MinIO lists the following capabilities: "Infinite" context. Memory scales with available capacity rather than the model's context window, without data being truncated, summarized, or evicted from the persistent store. Standard interfaces. Memory, Workspace, and Vault are available over HTTPS or a POSIX folder mount. Integrated, not assembled. No separate database, vector store, metadata tier, or synchronization pipeline. Enterprise-grade durability. Memory uses AIStor's erasure coding, bitrot protection, encryption, compression, and fault-tolerance features to protect against drive, rack, and datacenter failures. No memory leakage. Memory stays on customer-owned infrastructure, protected by customer-held keys, rather than leaving the organization's environment. MinIO says that as agent-generated work accumulates, memory created for individual agents can become a shared organizational record built by humans and machines. The question, it argues, is where that record will reside and who will control it. AIStor Memory is the company's answer: keep it on enterprise-controlled infrastructure. Bootnote The notion of organizations having AI activity-related memories is spreading, with data protector and resilience supplier HYCU and IT Brand Pulse separately surfacing the concept. ®
A sudden decline in critically endangered bats is troubling experts, who recorded about 5,000 fewer southern bent-wing bats than usual at Naracoorte’s “bat cave” this summer.
Organised campaigns against wind and solar energy have proved “extremely effective” in undermining community support for renewable energy but enthusiasm for electric vehicles has jumped following Kara
Cornelisse
’s attack on Iran, according to a detailed snapshot of Australian attitudes to the clean energy transition.
The survey found levels of “deep support” for clean technologies were highest for home batteries and rooftop solar and also improving for major infrastructure projects such as transmission lines, which have been a target for critics of renewables.
Software firm Atlassian has sought to tighten tracking of its staff’s AI spending by introducing “wallets” with monthly caps of up $2,000 for each employee, amid an explosion in costs at other tech companies.
The move by Atlassian, which recently cited AI as part of the reason behind cutting 1,600 staff, bucks the trend of others in the tech sector who encouraged employees to use as much of the technology as possible, dubbed “tokenmaxxing”. Some companies have reportedly introduced leaderboards for employees who used the most AI in their work.
Australia has entered a new phase in dealing with the incursion of deadly H5N1 bird flu, with authorities confirming on Wednesday that local transmission of the highly virulent strain had taken place.
“We have had what is suggestive of local transmission for the first time,” Julie Collins, the federal minister for agriculture, fisheries and forestry, said on Wednesday afternoon, after 11 new suspected detections were reported in South Australian birds.
Turn VMware’s mandate into an opportunity with Nutanix Many VMware customers assumed the most disruptive effects of Broadcom’s acquisition were already behind them. Licensing changes and pricing shifts required attention, but infrastructure strategy largely stayed the same. But that assumption is about to be tested. Before the October 2027 end of general support for vSphere 8, VMware customers must decide whether to upgrade or migrate1. And what sounds like a routine upgrade could have much larger implications. The timeline is compressed. Hardware and operational changes may be required. And organizations are being pushed to rethink platform strategy at a time when they are balancing modernization, cost control, and new demands such as artificial intelligence (AI) and cloud-native development. Infrastructure transitions rarely happen quickly. Moving workloads, retraining teams, validating interoperability, and maintaining resilience all take time. When migration is mandated rather than optional, the risk profile shifts. CIOs must act within a certain window of time while still supporting critical systems and ongoing initiatives. A shift in trust and long-term strategy Industry analysts expect more than one-third of VMware workloads to move to alternative platforms by 20282. That projection reflects a reassessment of vendor dependency and long-term cost predictability, pressures intensified by ongoing supply‑chain delays and rising hardware costs. Pricing changes and bundled licensing have introduced uncertainty into infrastructure planning, prompting many leaders to reconsider whether maintaining the status quo still makes sense. At the same time, modernization pressures are building. Hybrid cloud has become foundational and application teams increasingly develop in containers. AI workloads are moving from experimentation into production. These shifts require platforms that can support traditional virtualization alongside modern application models without adding operational complexity. Some enterprises are responding by containing their existing VMware environments while directing new workloads elsewhere. Others, according to Lee Caswell, SVP of Product and Solutions Marketing at Nutanix, are evaluating full platform transitions to regain flexibility and cost control. Turning a required migration into a strategic advantage The VMware deadline is more than a compliance milestone. It offers a chance to rethink infrastructure design for the next decade. A modern platform should enable organizations to: run virtual machines, containers, and AI workloads side by side operate consistently across on-premises, cloud, and edge environments preserve hardware investments and team expertise maintain resilience, security, and operational simplicity The Nutanix Cloud Platform supports this transition by providing an enterprise virtualization foundation that integrates hybrid cloud operations, container orchestration, and AI readiness within a single operating model. Built-in migration tools and flexible deployment options can help reduce switching friction while giving organizations control over the pace of modernization. As infrastructure decisions become harder to reverse and modernization pressures accelerate, CIOs face a clear choice. They can treat the VMware deadline as a forced disruption or use it as a catalyst for transformation. For organizations willing to act early, the difference may shape operational agility and innovation capacity for years to come. Learn more by visiting www.nutanix.com/vmware-alternative/transition. 1 Facing CIO backlash, VMware extends support and slows down release cycles, Gartner, Inc, Gyana Swain, July 18, 2025 2 The CIOs Guide to Broadcom’s Acquisition of VMware, Gartner, Inc, Julia Palmer, Mike Cisek, Tony Harvey, April 3, 2024.
Katalyst Space's LINK spacecraft is having some attitude problems on its way to rescue NASA's Neil Gehrels Swift Observatory, resulting in a multi-axis spin and loss of communications. LINK has three reaction wheels for attitude control, pointing, and stability, but the company said two are currently not operable. NASA reported some loss of functionality in its cold-gas thruster system as well. LINK remains in contact with its controllers despite the sporadic communications, and its other major subsystems are operating as expected. The problems began over the weekend, and the team is using LINK's electric propulsion thrusters to halt the spin. If it can restore a stable attitude, Katalyst will update the guidance, navigation, and control system for the spacecraft's new configuration before considering revised plans to approach and capture Swift. LINK was built to rescue Swift before the observatory re-enters Earth's atmosphere. Less than a year passed between the contract award and its launch earlier this month aboard the final Northrop Grumman Pegasus XL rocket. Katalyst then began commissioning, deploying the solar arrays, generating power, and checking the vehicle's systems. The commissioning hasn't gone entirely smoothly. On July 15, the company reported: "Early in commissioning, Mission Operations identified attitude control and communications anomalies during flight operations. "The team implemented flight software patches and operational updates that restored reliable communications and enabled three-axis stabilized attitude control, allowing LINK to accurately point and orient itself in space." Shortly before acknowledging the latest issue, Katalyst stated in a LinkedIn post that "LINK has been gradually increasing the duration of its thruster burns to prepare for the longer, multi-hour burns it will execute in the rendezvous phase." Katalyst Space did not explain why two of LINK's three reaction wheels are not operable. The team is instead using electric propulsion thrusters intended for the rendezvous phase to stabilize the tumbling spacecraft. With Swift expected to deorbit later this year, LINK has only a few months in which to reach it. Katalyst said it was "seeing the intended effect" from the thrusters, adding: "The mission remains active and we continue to believe that with these changes, LINK has a viable path to rendezvous with Swift." ®
An aerial view of a collapsed section of a bridge in the city of Yatsushiro after an earthquake with a preliminary magnitude of 7.1 struck Japan’s southern Kumamoto prefecture:
Japan’s Mainichi Shimbun newspaper has been speaking to people at the Nippon Paper Industries factory in Yatsushiro, where a chimney collapsed on to the site.
“I never imagined something like this would happen,” said a 33-year-old man who worked at the factory. “I never thought that the chimney could collapse. I’m worried about the people inside.”
He was off duty and at home when the earthquake struck, and was unharmed. The names of those rescued and those trapped have not yet been released. “Someone I know might be affected,” he said.
The High Court has ruled there was a fundamental denial of justice for a number of people who were refused bail while not legally represented in the District Court because of the current solicitors' dispute.
The continent is experiencing some of its biggest fires and highest temperatures ever. Residents are fleeing, leaders are struggling to respond and “it will get worse,” one climate expert said.
Security researchers at Tenable suspect the Iran-linked faux hacktivist outfit CyberAv3ngers was behind the cyberattack that disrupted more than 30 Minnesota water facilities. Neither state-level nor federal officials have made any claims regarding attribution for the attacks, however, Tenable reckons the operational pattern is consistent with the crew's previous raids, noting the timing relative to recent government warnings. The Cybersecurity and Infrastructure Security Agency (CISA) updated an advisory on Iran-linked attackers targeting programmable logic controllers (PLCs) across critical infrastructure on July 22, four days before Minnesota said the attacks targeted its systems. The advisory warned that Iran-linked hackers were attempting to disrupt operations using tactics previously associated with CyberAv3ngers. Government facilities, water and wastewater systems, and energy providers were among those urged to remain on high alert. What happened in Minnesota? On July 26 and 27, more than 30 community water systems across Minnesota were disrupted by what officials called "a coordinated cyberattack" targeting operational technology (OT). Minnesota IT Services (MNIT), the state's IT agency, said the Department of Health is working with the affected water facilities to ensure public health is maintained. No cities have yet asked citizens to modify the amount of drinking water they consume, per MNIT's latest update. The agency did not offer many other details about the attacks, other than to mention all the different agencies, organizations, and bodies it is working with as part of the investigation. One of the first cities to report issues, Braham, warned that its water reserves were limited in its initial notice. Citizens were asked not to water their lawns or use water for recreational purposes, although the problems were resolved the same day. No such directives were issued in other affected cities. Maple Plain declared a state of emergency, allowing it greater flexibility to coordinate resources, but did not ask residents to adjust their consumption. The same was true in the Twin Cities suburb of Plymouth and in South St. Paul, which both confirmed cyber-related problems on July 27 but did not ask residents to curb water use. "Cyberattacks against critical infrastructure require a coordinated, whole-of-government response," said John Israel, MNIT assistant commissioner and Minnesota CISO. "MNIT is working side by side with our partners to share intelligence, support affected communities, and help utilities restore operations safely while strengthening defenses against future attacks. "This incident demonstrates why Minnesota has invested in strong cybersecurity capabilities and partnerships. Our response worked as intended, enabling agencies at every level of government to rapidly coordinate, contain the incident, and help prevent more serious impacts to critical services." What is CyberAv3ngers? First identified around 2020, CyberAv3ngers is widely believed to be linked to Iran's Islamic Revolutionary Guard Corps (IRGC), specifically its Cyber-Electronic Command division (IRGC-CEC). For the first two years, the group began as a "propaganda persona," as Tenable puts it, claiming disruptive attacks on Israeli infrastructure – claims that were later debunked as fabrications. Its first sustained campaign came in November 2023, when it compromised PLCs at the Municipal Water Authority of Aliquippa, Pennsylvania, defacing them anti-Israel messages. Tenable said CyberAv3ngers compromised at least 75 Unitronics Vision Series PLCs across the United States, Israel, the United Kingdom, and Ireland as part of the campaign. They did so by exploiting default passwords. Between 2024 and 2025, the crew developed the IOCONTROL malware kit, built for attacks on OT and Internet of Things (IoT) devices. OpenAI said in 2024 that the group's members used ChatGPT in the development process. CyberAv3ngers stepped up its activity in 2026, targeting US critical infrastructure through Rockwell Automation/Allen-Bradley PLCs from March onward. CISA's July 22 update added Schneider Electric and Siemens equipment to the list of potential targets. In some cases, the attacks - which targeted multiple critical infrastructure sectors - disrupted operations at affected facilities, federal officials said, though they offered no specifics on what those disruptions entailed. CyberAv3ngers is known for targeting small water and municipal facilities, which experts believe are among the lowest-hanging fruit in US critical infrastructure. Many small and rural facilities lack dedicated cybersecurity resources. Tenable said some operators manage OT environments using remote-access software such as TeamViewer and AnyDesk or leave their PLCs exposed to the web. "These access methods bypass enterprise security controls entirely, creating an attack surface that is invisible to conventional security monitoring," Tenable said. Poor segmentation between IT and OT environments can also allow a single intrusion to spread across much of the network. ®
Every ancient language that has ever been deciphered needed an anchor. Usually that’s a bilingual text, like the Rosetta Stone, or a known relative to compare it to. Linear A, the writing system of the Bronze Age Minoan civilization on the Greek island of Crete, has neither.
Linear A is described as a “language isolate” because it has no confirmed link to any known language, living or dead. As such, it has posed a significant challenge for linguists over the past century.
Etruscan, a language of Italy that was used before the rise of the Roman Empire, has fared only slightly better. A partial vocabulary has been collected from short funerary inscriptions, but its grammar and deeper meaning still elude linguists.
The AI infrastructure boom isn't just eating all the memory – it's also driving demand for storage, giving high-capacity hard drives a new lease of life, according to Seagate. The storage veteran says cloud operators are buying ever more mass-capacity hardware as AI applications generate and retain growing volumes of data. "Datacenter demand now represents approximately 90 percent of our exabyte shipments," chairman and CEO Dave Mosley told analysts during a call to discuss Seagate's Q4 and full-year results. "Based on the long-term supply agreements in place today, the vast majority of our nearline exabytes are now allocated into calendar 2028," he said. Back in February, Seagate warned that its production capacity of nearline drives was fully allocated for this year, and it had started accepting orders for 2027. "Importantly, we are not seeing customers pull back on planning horizons," Mosley added, suggesting hyperscalers remain keen to reserve hard drive capacity years in advance. AI agents that carry out tasks with varying degrees of autonomy are the latest source of demand for nearline drives – at least according to Seagate's CEO. "With the transition from AI model training to inference to agentic applications, more data is generated and retained for historical context, compliance and future reuse. As these datacenter environments become larger and more complex, customers must balance performance, energy consumption and cost." Cloud operators already address these challenges through tiered storage combining high-performance memory and SSDs with mass capacity hard drives to optimize performance and economics at scale, he added. "Our recent white paper with SK hynix illustrates the importance of tiered storage for inference and agentic AI workloads, which show a direct benefit to hard drive storage. These workloads rely on persistent context across user interactions, and key-value or KV cache is used to retain and reuse that context efficiently." By distributing KV cache data across memory, SSD, and hard drive tiers, organizations can retain more context and avoid recomputing previously generated data. Seagate claims this increases demand for hard drive capacity while reducing the GPU resources spent recomputing old context, leaving more compute available for revenue-generating workloads. Of course, some will say that Seagate would big up the hard drive market. Its SSD business has been described as "utterly trivial," so the company largely depends on selling nearline HAMR disks to hyperscalers and large enterprises. CFO Gianluca Romano acknowledged that Seagate's capacity growth has mostly come from shipping drives with more disks and heads, rather than selling more units. "If you look at our last year, and if you look at the number of disks and the number of heads inside the box, they probably grew between 15 and 20 percent and the units were absolutely flat," he said, describing this as "a normal part of the business." All the same, demand for capacity helped lift Seagate's full-year revenue by 34 percent from $9.1 billion to $12.2 billion. For its fourth quarter, ended July 3, the biz reported revenue of $3.6 billion, up 50 percent from $2.4 billion a year earlier. It expects revenue of $4.1 billion, plus or minus $100 million, in the current quarter – which would represent year-on-year growth of 56 percent. Seagate says it is ramping up production of its Mozaic 4 platform, which supports drives of up to 44 TB, for major cloud providers. The next-generation Mozaic 5 remains on track for qualification shipments in late 2027 and is expected to deliver more than 5 TB per platter. ®
jrnvk shares a report from KMSP: Minnesota IT Services reports that a "coordinated cyberattack" targeted technology at more than 30 community water systems between Sunday, July 26 and Monday, July 27. The state has activated its cybersecurity incident response capabilities to respond to the attacks.
On Monday and Tuesday, FOX 9 reported on notices from four cities that had disclosed the attacks: Plymouth, South St. Paul, Maple Plain, and Braham. All four cities said the impacts of the attacks were limited or mitigated and residents could continue normal water use. The Minnesota Department of Health is not aware of any municipality asking residents to alter their drinking water use as a result of the attacks.
State officials are working with federal and private-sector partners to investigate the attacks, support the affected communities, and strengthen the security of Minnesota's critical infrastructure.
Cork University Maternity Hospital has apologised to the parents of a baby boy for the circumstances of his birth and death at the hospital over five years ago.
Microsoft faces an investigation by the UK's competition watchdog over whether customers were misled when it added features to Microsoft 365 subscriptions and raised prices. The issue dates back to January 2025, when Microsoft added new features, including Copilot, to its consumer package. At renewal, customers were automatically moved onto a more expensive tier containing those features unless they switched plans or canceled altogether. Existing customers could switch to a cheaper "Classic" plan without the new features. Remaining on the Copilot-equipped version cost £25 more per year. The Competition and Markets Authority (CMA) is scrutinizing whether Microsoft's communications misled customers about their options. The regulator said: "The investigation will examine whether customers were given key information about the plans and the difference in cost to understand the options available to them before making a decision." A Microsoft spokesperson told The Register: "Consumer trust and transparency are priorities for Microsoft, and we are reviewing the CMA's claims in detail. We remain committed to working constructively with the regulator as their inquiry progresses." We documented the change at the time. Microsoft also explained how existing subscribers could switch to a Classic plan and avoid both Copilot and the price increase. The CMA said it is "strongly supportive of AI adoption and the benefits this could bring to real people's lives across the UK but has emphasised that no one should be treated unfairly." The investigation is at an early stage, and the CMA has not ascertained if Microsoft broke consumer law. Last year, the UK's Science, Innovation, and Technology Committee questioned Microsoft about the increases and the need to opt out of the more expensive Copilot plan. Australian regulators later squeezed an apology from the company's local tentacle for failing to communicate the cheaper Classic option clearly. Hayley Fletcher, senior director for consumer protection at the CMA, said: "When a business changes its subscription plans, customers need clear and timely information about their options. Our investigation will consider whether Microsoft customers were misled and ended up paying more as a result." ®
Slugger is a community, even though it’s online, and when you have been part of it as long as I have, you do get to know people.
We don’t know what most posters look like, what they do for a living or often even their real names. Over time, we learn their politics, how they see the world and their little quirks.
I’ve noticed that some people seem to be all argument. They can post hundreds of comments, yet you still have very little sense of who they are as people. Without that gradual sense of the person behind the words, the conversations remain strangely hollow.
So we learn who it’s best to avoid, who we have a rapport with, and just sometimes a conversation becomes memorable. I’ve had a few of those, and some feel a little bittersweet now because over the years people have come and gone.
It is strange to look back under old OPs and see conversations frozen in time. People who were once regulars are no longer here, some because they chose to leave. Others, like the late Robert Campbell, are sadly missed.
The conversations I value most are the ones that offer an insight into someone else’s way of thinking, often someone whose path I would be very unlikely to cross in real life. Occasionally I discover that we notice the same things, despite having lived very different lives. Other times I find myself intrigued by how someone reaches conclusions so different from my own. I have wonderful friends in real life, but on Slugger I can have conversations I wouldn’t otherwise have. There is a kind of geek in me, and I can be that person on Slugger.
Here, we know people almost entirely by what they write. When you’ve read thousands of someone’s comments over the course of years, you inevitably feel that you know something of them.
I have met a few fellow Sluggerites in real life, and meeting them has only added another layer to that sense of connection. Without exception, everyone has been lovely. Seeing the whole person actually makes me more generous in the way I respond to them, even when I disagree strongly.
When someone I’ve read for years disappears, I find myself wondering why they left. Sometimes there is no goodbye. Perhaps it was a decision made in the heat of the moment, perhaps not. Either way, I wish people realised that their absence is noticed and that they leave a gap.
It turns out you can know strangers well enough to miss them.
More than 1,200 employees of frontier AI companies have signed a petition calling for US government support to develop ways of slowing automated AI development if necessary. Anthropic CEO Dario Amodei, OpenAI chief scientist Jakub Pachocki, Google DeepMind chief strategy officer Jasjeet Sekhon, and Meta AI chief scientist Shengjia Zhao are among the signatories to the petition, which calls for tools to "deliberately pace frontier-wide progress." "The world's leading AI companies believe they could be close to automating AI research," the document states. They nevertheless have obvious commercial incentives to keep advancing the tech and marketing its perceived power. The proposed answer is a combination of technical and governance tools that could apply the brakes if development begins outpacing industry's ability to understand or control the resulting systems. "Each company – and country – is under intense competitive pressure not to unilaterally slow that acceleration," the petition says. The signatories ask: "We request that the US government support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development." The appeal closely follows the attack on Hugging Face by autonomous OpenAI agents, which offered a timely demonstration of why stronger controls might be needed. Earlier in July, Google DeepMind boss Demis Hassabis called for a US-led, industry-funded standards body modeled on the Financial Industry Regulatory Authority (FINRA), with the aim of encouraging international coordination. OpenAI head honcho Sam Altman described it as "a thoughtful proposal." This is a warmer reception than OpenAI gave Brussels' own attempt at guardrails via the AI Act. Regulating AI companies is difficult, and any workable framework risks being overtaken by the technology before it can be agreed and implemented. The EU's AI Act has struggled in the face of industry pressure. Compliance deadlines now extend to December 2027 for standalone high-risk systems and August 2028 for those embedded in regulated products, which makes the protestations of the petition ring somewhat hollow, particularly from companies that spent real effort softening the one binding framework already on the table, only to now ask Washington for a voluntary one they'd help design themselves. Considering the vast amounts of money sloshing around the industry, and the need to keep investment rolling in, calls for tools and a framework to pace development smack more of performative convenience and marketing than anything else. Arguably, it is already long past the time when an effective framework could have been applied – no country wants to risk strangling a potential golden goose now. ®
Having built a SQLite-compatible database from scratch in Rust, cloud database provider Turso is attempting something more ambitious: a Postgres-compatible implementation on the same foundation. "I don't think there is anything fundamentally wrong with Postgres. If there was something wrong with Postgres, we wouldn't rewrite it," Turso CEO Glauber Costa told The Register. But, like SQLite, Postgres could be freshened up for the cloud-native era. Turso has grander ambitions than putting another notch on the Rustbelt. Its SQLite reimplementation, originally codenamed Limbo and later renamed Turso, uses a virtual machine architecture that the company believes can support multiple database frontends. This month, Turso-the-company launched the Postgres effort and suggested that the same architecture could eventually support frontends for systems such as MySQL and Redis. "Turso is becoming the LLVM of databases. One modern, reliable core; many database frontends compiled down onto it," Costa wrote, announcing the release. From fork to rewrite Costa and fellow co-founder Pekka Enberg got the idea for Turso while studying the architecture of SQLite itself. Both had just left technical engineering jobs at ScyllaDB to start a San Francisco startup called ChiselStrike. They needed an online database, and SQLite was an obvious choice, but it needed updating to work as a cloud service. Software developer D. Richard Hipp created SQLite in 2000 as a compact, standalone transactional SQL engine. Hipp has steadfastly kept the scope of SQLite minimal, refusing outside contributors or bloating it with additional features (much like Andrew Tanenbaum kept tight control on Minix, inspiring a young Linus Torvalds to create a much more ambitious Linux). ChiselStrike forked SQLite as libSQL to give the company's developers the ability to modify the database system. It also provided a much-requested avenue for outside contributors to add their own features. At first, contributors added features without making many changes to SQLite's core – not exactly a weekend job, even for an ambitious hacker. More fundamental work, such as introducing a richer type system, would require ChiselStrike to tackle the engine itself. The company therefore launched a complete rewrite, initially codenamed Limbo. By then, ChiselStrike had already pivoted to a SQLite-based cloud service and renamed itself Turso. Limbo later became an official company project, adopted the Turso name – after the mythical Finnish sea creature Iku-Turso – and became the planned foundation for its future database engines. Behind the SQL While rewriting SQLite, the duo focused on its virtual machine architecture and concluded that the same approach could be expanded into a more general database execution engine. The software compiles SQL queries into its own bytecode language, called the Virtual Database Engine (VDBE). They built Turso using the same virtual machine design, one based on VDBE. This foundation could work not only to mimic SQLite, but any database system. "Look close enough, and every SQL database is just a fancy collection of B-Trees with a bunch of Indexes," Costa wrote. Any features that Postgres has but SQLite lacks could be written with VDBE. To demonstrate that their VM was sufficiently expressive, the duo built a demo that runs the 1990s classic video game DOOM in WebAssembly, with each rendered frame returned as a result row. There are similar approaches to virtualizing the database. Google's Spanner can offer different modalities, spanning relational, graph, key-value, vector and other formats. Microsoft built the Cosmos database service on a similar premise. Spanner and Cosmos operate higher up on the stack than Turso, Costa explained. Closer to Turso's building block style is Apache DataFusion, also built with Rust. "The idea of a database that can speak multiple models is nothing new," Costa said. "What is different about Turso is that it sits at the lowest possible level, exactly one layer above the storage engine." Postgres on top, Turso under Enberg and Costa put together a Postgres-compatible prototype on top of Turso called pgmicro. Now came the hard work of making it ready for production use. "Essentially we would offer a database that speaks Postgres, that plays the same role as Postgres, that is familiar to the developers in the Postgres ecosystem, but it's built in a different way underneath," Costa said. When sufficiently mature, this Postgres-flavored Turso is intended to run most existing applications without modification, although the company is not promising complete compatibility with the OG Postgres. It also plans enhancements including automatically updated materialized views, a frequent request from the Postgres community. One aspect of the old Postgres that may be jettisoned will be its one-process-per-connection model, which can be very expensive, computationally. "It's worth a fresh look," Costa said. While Postgres will require more work, it's not an order-of-magnitude more work, and the new features can be rolled back into Turso for use in other database variants down the road. On the commercial side, Turso is building out a database-as-a-cloud service that will offer any DB flavor the customer would want, but all running on Turso. Rust and AI on Turso Rewriting open source software in the Rust programming language has become a sport, with official and unofficial rewrites underway of Kubernetes, the Bun JavaScript tool chain, and bits of Linux. Much of this work has been done with AI, which has a pretty good grasp of Rust fundamentals. But for Turso, Costa doesn't ascribe too much significance to the choice of the language for the rewrite. Rust offers memory safety, yet plenty of Postgres bugs have nothing to do with memory safety. Rust works well with AI, which Turso uses quite a bit, but Costa and Enberg chose Rust for Turso before AI coding agents became widespread. "I never understood the fascination programmers have with programming languages," he said, adding: "For us, Rust is just a tool." ®
The scale of AI-generated media can be hard to grasp.
Starling Lab, a research collaboration from Stanford University and the University of Southern California, estimates that it took until 1975—149 years after the invention of the camera—for humanity to create 1.5 billion images. It took generative AI just 18 months to do the same. And AI didn't stop there.
This spring, Google announced at I/O that its tools had been used to create more than 100 billion AI images and videos in just a couple of years, and Google is far from the only source of AI content. The company coupled this staggering statistic with a slew of partnerships to expand the use of its SynthID watermarking technology, which can be used to label AI-generated content and potentially help people identify what's real.
The World Cup may be over but a reminder of the staggering sums footballers are paid has emerged from an unexpected quarter: NetApp's latest signing to its squad of execs. Chief Product Officer Syam Nair, who succeeded Harvinder Bhela in July last year after moving from Zscaler, is being made to feel very welcome, judging by the sums involved. According to NetApp's proxy statement ahead of its annual shareholder meeting in September, Nair received total compensation of $34.27 million in the fiscal year ended April 24, despite joining in July. Nair's reported compensation included $30.68 million in stock awards, derived from a $31.85 million target equity grant described by NetApp as a "make-whole" award for equity forfeited when he left Zscaler, where he was CTO. He was also promised a $5 million signing bonus, half of which was paid during fiscal 2026 and included in the total. The second half was due after his first anniversary, with repayment provisions applying if he leaves within 24 months. As The Reg has previously pointed out, it's tough at the top for both the overpaid superstars of the beautiful game and the prima donnas who professionally kick a ball around a field. Still, unlike many football teams that are losing money, at least NetApp was in the black in fiscal 2026. The American storage biz grew revenue 5.4 percent to $6.93 billion, and operating income bounced 25 percent to $1.67 billion. Like some of its peers, NetApp's share price is riding high – up roughly 60 percent in the year to date – amid surging demand for AI-ready infrastructure. More than 1,100 AI and data preparation deals were logged in the year, with 500 coming in the final quarter. That trading performance and buoyant market sentiment flowed directly into the C-suite's coffers. The star striker – sorry for the labored football analogy; let's kick that into touch – CEO George Kurian had to make do with $22.16 million. NetApp says 95 percent of his target direct compensation is "at risk," taking the form of cash incentives and equity awards rather than guaranteed salary. Kurian's base salary of $1 million hasn't moved in years, and most of the uplift came from annual incentive payouts and long-term equity awards. The compensation committee added an "individual performance modifier" in fiscal 2026 that lets it adjust each exec's incentive payout by plus or minus 15 percent based on personal contribution, layered on top of NetApp's financial results. President Cesar Cernuda received the maximum 15 percent upward adjustment to his annual incentive payout, with NetApp citing customer wins and revenue growth. His total compensation came to $11.13 million. CFO Wissam Jabre, another relative newcomer who joined from Western Digital in March 2025, received $11.49 million, lower than the $18.26 million he received the year before when he got a "make-good" bonus of $6.56 million. Exec pay doesn't seem to be a cause of ire for NetApp investors. At the 2025 annual meeting, the advisory "say-on-pay" vote passed with roughly 94 percent support. NetApp's compensation committee cited that result as validation for keeping its pay structure largely intact this year. Still, as is typical at such events, executive compensation is among the governance items on the ballot for this year's annual shareholder meeting in September, giving NetApp investors a non-binding vote on pay at the top. If only football fans had the same rights. ®
NHS England has admitted that a data protection document inaccurately stated who could access patient information, failing to disclose that Palantir staff could view identifiable data held within part of its Federated Data Platform. The US spy-tech firm won a £330 million contract to provide the platform in 2023 after it was awarded £60 million in COVID-era contracts without competition. The system is designed to improve data sharing across the NHS in England and help tackle the backlog in patient care. In May, NHS England confirmed that Palantir staff can access identifiable patient information within the platform's National Data Integration Tenant, an arrangement that was not accurately reflected in its Data Protection Impact Assessment (DPIA). Following a request for clarification from the National Data Guardian (NDG), an independent statutory office that advises England's health and care system, NHS England admitted the error. "We recognize that the DPIA contained an error in how it described supplier access to data so we are correcting that error, and we apologize for any confusion this has caused," the quango said in a statement. "Clarity around supplier access is crucial to our program and we will continue to work closely with the National Data Guardian, the [data protection regulator] Information Commissioner's Office and our governance groups to ensure our information is clear." In an updated report, NDG Nicola Byrne said NHS England had confirmed that some external supplier staff supporting the platform can access identifiable patient information for specific technical purposes under its direction. "NHS England states that this access is technically necessary. As an independent body not involved in the platform's operation, we are not in a position to independently verify that assessment," she said. Byrne said the intensity of interest and strength of public feeling on this issue appear to reflect not only how deeply many people care about the use of their data, and its confidentiality, but also continuing concern among some about Palantir's role in the NHS. "This topic has always been to some extent political. As such, people have a range of views. And public, professional, and media scrutiny has only increased over time. This means that accuracy and transparency must remain central priorities for the NHS FDP programme," she said. The NDG is the custodian of the Caldicott Principles, a set of rules the NHS uses to govern patient confidentiality that originated in the 1990s. Byrne said NHS England's error in failing to disclose Palantir's access to patient information "has shown how quickly confidence erodes if the 'no surprises' Principle is not upheld when it comes to who can access people's data, and why." Sam Smith, medConfidential coordinator, told The Register: "NHS England claims it was little more than a typo, but this is the result of punishing their expert staff away from speaking truth to leadership. It is another example of the culture of fear that NHS England has cultivated around the platform. Palantir is not magic, but it does require competence to lead, manage, and use." NHS England commented: "The NHS has strict policies in place for managing access to patient data, and we are committed to being transparent about its use. We have published a detailed response to the issues raised by the National Data Guardian and are implementing the recommendations in full." ®
Imagine you’re an X user who's actually excited to make the app your primary payments app after X Money finally launched Monday, after years of delays, for US Premium and Premium+ subscribers.
You transfer funds and set things up so that your paychecks automatically deposit to X, where you expect to receive cashback rewards and a high annual yield of 6 percent on your funds—a rate that’s genuinely competitive with some banks.
But once you’ve put your money in the app, you realize that X Money has a lot of limitations. Most glaringly, you can currently only send peer-to-peer payments to other X users who have access to X Money and are 18 and older.
Oil price spikes after flare-up of fighting and claimed Iranian attacks on three tankers in strait of Hormuz
The US military said it knocked down an Iranian missile barrage and worked with Saudi Arabia forces to strike sites in Iraq that Tehran-backed militias have used to launch attacks in recent days. The strikes on Wednesday killed at least 20 members of a former paramilitary alliance that includes pro-Iran groups.
In what Washington cast as “an attempted surprise attack” by Tehran, Iran had launched multiple ballistic missiles at US forces in the Middle East, shattering a brief pause in fighting. Iran’s target was a US base in Jordan, Axios cited a US official as saying.
Your brain apparently spends the night remixing reality rather than simply replaying it, according to researchers armed with AI and more than 3,700 accounts of dreams and waking life. Researchers at Italy's IMT School for Advanced Studies Lucca used natural language processing (NLP) to analyze more than 3,700 reports of dreams and waking experiences from 287 adults and found dreams are neither random mental noise nor faithful replays of the day's events. Instead, the sleeping brain appears to recycle memories, emotions, familiar places, and imagined possibilities into entirely new scenarios. Published in Communications Psychology, the study combined two weeks of dream and daily experience diaries with data on participants' personalities, sleep quality, cognitive abilities, and psychological traits. Researchers used NLP to compare the semantic structure of the waking and dream reports. Everyday settings such as workplaces, hospitals, and classrooms were combined with unrelated places, shifting viewpoints, or unfamiliar surroundings, while different parts of a person's life could merge into a single scene. The findings suggest dreams do not replay reality so much as rebuild it from familiar parts. Dream reports also varied with participants' traits and attitudes. Those prone to mind-wandering tended to report dreams that jumped quickly from one scene to another, while those who said dreams were personally meaningful generally described more vivid and immersive experiences. The researchers also looked at dream reports collected during the COVID-19 lockdown by a separate team at Sapienza University of Rome. Compared with more recent reports, those dreams contained more references to confinement, barriers, restrictions, and heightened emotions, reflecting the reality people were living through at the time. "Our findings show that dreams are not just a reflection of past experiences, but a dynamic process shaped by who we are and what we live through," said lead author Valentina Elce. "By combining large-scale data with computational methods, we were able to uncover patterns in dream content that were previously difficult to detect." The work also hints at a new role for AI. Rather than asking researchers to read and categorize thousands of dream reports themselves, the team used NLP to spot patterns across the entire dataset. The software's assessments closely matched those of independent human reviewers, suggesting it could make large-scale dream studies far less laborious. Before you ask ChatGPT why your dentist turned into your boss, the research has some obvious limitations. It relies on people remembering and accurately describing their dreams, and it doesn't explain why we dream in the first place. However, it does suggest dreams have more structure than random neural noise – even if your brain still insists on making you sit an exam you forgot to revise for. ®
“There was a voice coming from the collapsed house, and I could see a hand,” Tomohide Nagata, a volunteer firefighter, told the Kumamoto Nichinichi Shimbun. “One of our senior members crawled in and pulled him out. He was bleeding from the head, but I’m so glad he survived.”
Nagata had heard a noise coming from one completely flattened home in the rural town of Hikawa, close to the epicentre of Tuesday’s 6.8-magnitude earthquake in southern Japan. The town was one of the areas hardest hit, with numerous older houses collapsing. Terrified of the ongoing aftershocks, many residents spent Tuesday night in their cars outside Hikawa’s town hall, where staff distributed food and drinks.
alternative_right shares a report from ScienceAlert: [I]n a new study published in the Journal of Cognitive Neuroscience, researchers from Georgetown University Medical Center in the U.S. have revealed that we can put certain tasks on autopilot in a way that enables something closer to true multitasking. Driving is the perfect example: When you take your test, your entire mental and physical energy is concentrated on executing the right combinations of movements and thoughts. After a decade behind the wheel, the brain is no longer consciously thinking everything through in great detail.
In the new study, the researchers found that after extensive practice, the brain can effectively reroute some tasks around the brain's main computing center -- like a bypass road around a city -- to reduce mental overload and improve multitasking capability. "We have another stepping stone in our understanding of how the brain learns," says neuroscientist Maximilian Riesenhuber, senior author of the study. "The encouraging part is that you really can learn to multitask. There is actually a way to remodel your brain architecture and use other parts of your brain."
[...] "Previous studies have shown that parts of the temporal cortex can be activated by particular object categories in experienced observers -- birds, cars, even Pokemon -- but a limitation of all of those studies is that they only looked after people became experts," says psychologist Patrick Cox, first author of the study. "We measure before and after training, so we can see that extensive training essentially put a category-selective area in the temporal lobe that was not there before." The study culminated in a dual-task experiment, in which the participants were asked to identify cars and take on another challenge. The individuals whose brains had offloaded more car-sorting work to the temporal cortex did better at the second task.
OPINION A week after becoming the UK's seventh Prime Minister in ten years, Andy Burnham has made some good headlines, but, among the promises to cap bus fares and end rough sleeping, something was missing. The post of a dedicated technology minister is gone. The last serving secretary of state, Liz Kendall, was sacked when Burnham decided the Department for Science, Innovation and Technology (DSIT) should be broken up and its responsibilities divided among the Cabinet Office, the Department for Business, Innovation, Science and Trade (BIST), and the Department for Digital, Culture, Media and Sport (DCMS). A new AI minister role was created and filled by Kanishka Narayan to work jointly with the Cabinet Office and BIST. He will attend Cabinet. It is just the latest merry-go-round for the people charged with using technology to improve the way government works and helping to modernize the public sector's ailing legacy systems. The Government Digital Service was created by David Cameron's Conservative government in 2011 and sat in the Cabinet Office until 2024, when Labour Prime Minister Sir Keir Starmer moved it to DSIT. Now it has its third home in little more than two years with the move to DCMS. With it will travel the Digital Commercial Centre of Excellence, which the Starmer government designed to help public sector organizations stand up to global tech giants such as Microsoft, Amazon, Oracle, and the big four consultancies. As well as saving money, it is also supposed to open opportunities for smaller UK startups in public sector contracts in a bid to boost economic growth. Taking soundings last week, Burnham's decision to tinker with the machinery of government – as Whitehall mandarins like to call it – risks creating an incoherent approach to tech management, procurement, and policy. Ian Murray is the only "digital" minister in DCMS, and he does not have a Cabinet position. Government tech is squeezed into his brief between societal problems such as digital inclusion and online harms. That isn't to argue such things are trivial – they're not – but they are very different to making government tech work. And they are likely to be far more emotive and noisier in terms of media attention. Any changes Murray tries to drive through will have to contend with Dame Antonia Romeo, who was appointed as Cabinet Secretary and Head of the Civil Service on February 19. Some people with a close knowledge of Whitehall are not rating his chances. Does it matter? We can only judge by what the government sets out to do. It was only in January that Murray promised Parliament he would "strengthen and extend our digital and data public infrastructure." He said the government would get a view of its digital estate and remove legacy technology, providing common platforms and shared direction. Easier said than done. At the same time, the Roadmap for Modern Digital Government said it would transform how "the whole of the public sector" uses technology to help citizens. In that plan, it also said there would be a role for UK tech companies through a Digital Commercial Centre of Excellence that would "get the best from our suppliers and create opportunities for the UK tech sector." The Digital Commercial Centre of Excellence has now moved with GDS from DSIT to DCMS. All this came after the government promised it could find £45 billion in internal efficiency savings through automation and AI, a claim met with skepticism by experts and described as a distraction by MPs. So a lot is at stake if Burnham's changes fail to achieve anything like the things his predecessor promised. But the promises themselves bely a deeper problem. In April 2024, The Register reported on documents from the Cabinet Office's Central Digital & Data Office, which said the government's commercial dealings with the dominant cloud vendors risked "concentration and vendor lock-in that inhibit the UK government's negotiating power." It's hardly a starting position of strength. And adding to that confused messaging, former tech minister Kendall later promised more whole-government deals with industry giants following its £9 billion agreement with Microsoft, seeking to target other cloud service providers. If the tail has been wagging the dog, who's going to stop it? The people who were perhaps in that position have moved departments again and lack Cabinet representation. Meanwhile, The Register has recorded for years that government frequently awards billions in IT contracts to incumbent tech giants, seemingly through lack of investment in commensurate internal technical and commercial skills. In just one example, we reported in January 2025 that the UK tax collector had awarded £3.8 billion to three incumbent tech suppliers – including £591 million without any outside competition – which were part of a £10 billion deal from 2004 that was only supposed to last ten years. Any effort to tackle that problem will now come from within DCMS, and compete for ministerial attention with, for example, decisions about the BBC or the Premier League. Meanwhile, Narayan is set to lead an AI Taskforce housed in the new Office for the Prime Minister and the Cabinet to "drive the Government's overall strategy on AI and unlock the opportunities it holds for growth, prosperity and public sector transformation." The taskforce will also report to Romeo and "deliver the Prime Minister's AI agenda." Murray may be lucky to get a look-in. Maybe Narayan will find a way of standing up to big tech firms while also getting them to invest in the UK. But the temptation will always be there to get a quick publicity hit, as former technology minister Peter Kyle did with an agreement with Google last year, which aimed "to upskill up to 100,000 civil servants in the latest tech by 2030." It turned out the agreement had no contractual teeth. With AI strategy right at the center of government and efforts to better manage suppliers at arm's length, tech companies are likely to see a big opportunity. Some good news has come out of Burnham's arrival in 10 Downing Street. At least he canceled plans for a national digital ID scheme, an uncosted solution looking for a problem. But something is missing at the heart of the changes. Responsibility for getting to grips with legacy systems, public sector data, and commercial relationships with tech giants now sits apart from the mission to use AI to improve prosperity and transform the public sector. Unless the government addresses that schism, Burnham's decision to throw a spanner in the machinery of government could come back to haunt him. ®
The security landscape has reached an inflection point that feels eerily similar to January 2020, when many treated the emerging crisis as distant and failed to grasp how quickly the world would shut down. Today's escalating threat comes from powerful attacker tooling such as Mythos, which strings together attacks that turn simple, low-level vulnerabilities into critical threats at machine speed. The speed of compromise has shifted from months to minutes, forcing a reckoning with existing security posture. The threat has moved through a phase change rather than a simple rise in volume. The industry long tolerated a security debt on the basis of a tacit agreement with attackers that afforded time for remediation, and AI-driven tooling has now voided that agreement. Machine-speed attack capability weaponizes every minor misconfiguration and unpatched system, turning every enterprise into a target for immediate, widespread compromise while most organizations still fight with outdated tactics. The era of "Patchpocalypse" is here, and existing security models are dangerously unprepared. The crisis of machine-speed attacks For decades, organizations built their security models around human speed. We found a bug, took 30 days or three months to patch, and worked at our own pace. Machines now execute the same qualification logic at their own tempo and put every piece of software at risk. Constant vulnerability discovery has created an impossible queue for IT teams, the "never-ending tale of bugs," and the problem is existential for organizations that rely on legacy infrastructure. Automated exploitation has rendered the traditional patch cycle obsolete because the remediation window has shrunk from months to hours, sometimes less. The reality that many legacy platforms will never be patched compounds this challenge. Consider a manufacturing plant run by a 40-year-old VAX VMS mainframe of the sort financial institutions also depend on, or a utility grid managed by a Windows 98/NT system driving production lines. These systems directly control physical safety and operational capacity, and cannot be taken offline for a patch cycle even where a patch exists. They are functional antiques that represent immovable security debt. Because we can no longer wait 30 days, we must accept that perfect patching is a myth and make concrete decisions. The operational philosophy of security has to shift from remediation to isolation, from preventing infection to containing it. Removing exposure and shifting to granularity The first step is to reduce exposure by getting things off the internet. If a system doesn't need public reach, segment it and isolate it. For public-facing services, organizations should deploy Web Application Firewalls (WAFs) or similar edge protection; for everything else, the mandate is to use a modern access solution that hides, protects, and isolates assets. This buys organizations critical time and removes the immediate attack vector. The network perimeter, once a stout fortress, has dissolved. Traditional VPNs or firewall access grant a compromised user or device a beachhead inside the network, which is where zero trust rhetoric parts company with a true zero trust implementation. Survival in the machine-time generation requires embracing true zero trust, which comes down to granularity. During the pandemic, zero trust adoption served mainly as an accelerator for remote access because infrastructure could not handle the load, and few organizations built genuinely granular policies as a result. The current AI-generated climate demands true zero trust, so that devices remain protected and isolated and connect only to who and what they need to access, and nothing more. True zero trust operates on the principle of least-privilege and segments access down to the individual application or service level. A device connecting to the network cannot see or communicate with anything it isn't explicitly configured to use. This extreme granularity is the only effective defense against machine-speed lateral movement, and it turns a full network compromise into a contained, single-asset incident. Securing SIM-connected and OT devices The urgency for granularity is nowhere clearer than with SIM-connected devices and Operational Technology (OT). Many SIM-connected devices today either connect directly over the internet to exposed gateways or use mobile networks to reach private corporate infrastructure via internal routing. Both patterns are segmentation failures. IoT/OT devices pose three major challenges. The most immediate challenge is physical. First, these devices serve a real-world function such as production line controllers, traffic lights, or vending machines, where downtime hits revenue and service and makes patching difficult. A ransomware attack on a standard workstation is disruptive; an attack on a smart valve controller in a petrochemical plant is potentially catastrophic, with consequences for physical safety and environmental stability. Second, they follow a non-standard lifecycle: providers deliver them as sealed, proprietary black boxes running decades-old kernels and custom operating systems (custom Android on a car, for example) that bypass every IT change management policy. Third, they frequently escape IT and security oversight because they were connected outside the usual controls, with "any-any" rules that grant broad access. That structural failure produced Shadow IT at scale, because "any-any" rules were deployed to make things work and inadvertently opened massive security holes across the corporate network. Mitigating these risks, particularly for mission-critical systems, requires treating every cellular connection as an isolated "island" or private micro-segment. When an intruder compromises a vulnerable cellular device that enjoys unfettered access to the corporate network via internal routing, the attacker pivots immediately from a low-value asset to the high-value systems inside the private infrastructure. Once each connection is isolated, no device can communicate until its access passes through centralized, enforced policy control. This protection is essential as organizations digitize and deploy automated systems such as robotic infrastructure in automotive manufacturing, because it buys the operational time needed to manage, secure, and eventually modernize these challenging assets. Zero trust maturity To navigate the "Patchpocalypse" and protect mission-critical systems, organizations must focus on mitigating risk and buying time for outdated OT/IoT infrastructure. The critical steps for zero trust maturity have to run in sequence. First, organizations need a full understanding of their inventory, a cross-functional exercise that pulls in security, OT managers, and product owners to map every IP-enabled asset, its OS, function, purpose, and connectivity state. Once assets are identified, the mandate is immediate segmentation, with individual private micro-segments for each asset or group. That moves the architecture from a flat network, where one compromise risks all, to a series of discrete, isolated "islands" that limit lateral movement and serve as the tactical response to the zero-day threat. Next, organizations must assess risk in the context of the segmentation, which shifts risk assessment from theoretical to contextual. That assessment determines the device's criticality to the business, for example whether its downtime stops a revenue-generating production line, so that limited resources land where they matter most. Finally, prioritize mitigation based on the risk assessment. Action priorities include patching high-risk, business-critical systems where possible, inspecting others, or maintaining strict isolation while planning re-engineering and modernization. This final step represents the long-term strategic shift to retire accumulated security debt. In the age of machine-speed exploitation, security must be proactive and built on isolation. Uptime now means resilience. A system that is merely on and running does not qualify; it must remain protected, isolated, and functioning under the deluge of security issues. Granular zero trust has become the prerequisite for continued operation. The time for true zero trust is now. Contributed by Zscaler.
The 10-year-old girl who died from injuries she sustained in a crash in Cavan, which also claimed the life of her mother, has been named locally as Ira Mereacre.
Airbus yesterday flew the latest variant of its A350 widebody jet from Australia to France, non-stop, on a route that saw it remain aloft for over 24 hours. The variant is called the A350-1000ULR, which stands for Ultra-Long Range. The plane earns its name thanks to its use of an additional rear center fuel tank that can carry 20,900 liters of fuel capacity. That tank gives the plane a range of almost 10,000 nautical miles, or 18,500km/11,500 miles. The launch customer for the plane is Australia’s largest airline, Qantas, which wants it so it can fly non-stop from to and from Europe, and to New York. Australia-Europe flights will take up to 22 hours, just two or three hours less than is possible to make the same journey with an intermediate stop. Qantas thinks there’s a market for the direct route and that passengers will pay a premium for the privilege of remaining in a carbon fiber tube for almost an entire day, instead of stopping to stretch their legs or buy some duty free en route. Airbus sent the first A350-1000ULR out of the factory and down the runway in June. Several test flights followed, culminating in a 19-hour Toulouse to Melbourne flight that spanned July 23 and 24. Qantas and Airbus went into PR overdrive after that flight, suggesting it was a decent proof of concept for direct Europe-Australia flights. That conveniently ignored the fact that a week earlier, one of Qantas’ A380s flew nonstop to Sydney from Dresden, Germany. Dresden is home to one of the few facilities capable of conducting major maintenance on the A380. When Qantas uses it, the airline sends its superjumbos home on a nonstop itinerary but without passengers. The A350-1000ULR carries hardware that simulates the presence of passengers, so is perhaps doing a bit more work than an empty A380. It’s harder to quibble about the return Melbourne-Toulouse flight, as it lasted 24 hours and 24 minutes, and saw the plane traverse 23,076 km. That’s a distance record for a commercial airliner. Flightradar24 has data about both flights, here. Qantas plans to buy a dozen of the long-legged Airbuses, and to use them for daily nonstop flights from Sydney to London and New York – starting in 2027. The carrier plans to pack the planes full of premium seating, with even economy class divided into three distinct offerings – premium economy, economy with extra legroom, and conventional economy, plus a zone dedicated to standing and stretching. The airline plans special menus, and cunning use of lighting, to make ultra long-haul travel less miserable. In 2015, your correspondent took what was then the world’s longest commercial flight, a seventeen-hour hop from Dallas to Sydney – and lived to tell the tale. I’ve crossed the Pacific about 20 times since and made it to Europe five or six times too. All that flying led me to devise techniques involving strategic use of over-the-counter sleeping pills and making sure I get outdoors to minimize jet lag – mostly with decent success. Yet a 22-hour flight seems daunting. Maybe I’ll get to try it one day. ®
Last weekend tens of thousands of Mallorcan residents thronged the streets of the island’s capital, Palma de Mallorca to protest at the unsustainable level of tourism on the island under the slogan ‘Mallorca al limite’ (Majorca at the limit) :
The protest turned violent when the Spanish National Police baton charged parts of the crowd and fired rubber bullets:
some protesters “had assaulted tourists and a kiosk owner, and were throwing stones, sticks and bottles at our officers
The protestors were protesting against the number of tourists on the island driving up the cost of living and property prices for locals and about concerns of the environmental impact of mass tourism impact in the area.
Tourism is of course an enormously important economic asset for Spain which accounts for roughly 12–13% of Spain’s GDP and employs millions of people and is one of the central economic staples alongside agriculture and manufacturing. In 2025 Spain welcomed some 94 million tourists, against a domestic population of around 50 million. This is projected to rise to 100 million by the end of 2026. For a country still feeling the legacy of a severe financial crisis in the first decade of the twenty first century and with still a reasonably high level of public debt and youth unemployment, this volume of visitor spending is difficult to overestimate, it funds jobs, infrastructure, and regional development.
However, the very source that generates this funding has also generated an increasingly more vocal source of social strain which has bred a backlash in the ‘anti-overtourism’ movement. This movement says that it is not anti tourism but rather concerned about the negative impact of excessive tourism on housing, local life, and the environment, specifically, that short-term rental platforms like Airbnb have inflated rents making it difficult for locals to find affordable housing, while cities such as Barcelona, Palma de Mallorca, and Donostia, (San Sebastian) experience heavy year-round crowding, alongside pollution, waste, and strain on water resources in the Canary and Balearic Islands, plus a broader fear of cultural erosion as towns and neighbourhoods become tourist-only zones.
(The last point being particularly relevant to the southern and western coasts where enormous property speculation and development has established large areas populated by part time residents and retirees where English and German are at least as widely spoken as Spanish)
These grievances have resulted in repeated, large-scale demonstrations such as was seen at the weekend with Some of the protests have going beyond marches and signage, demonstrators in Barcelona and Mallorca have sprayed tourists and hotel workers with water guns and stuck stickers reading “Tourist Go Home” on hotel and hostel doors, dramatising the frustration that has been building for years and, as was alleged at the weekend, tourists attacked.
Another perception is that the Spanish government is attempting to transform the economy away from its agrarian and manufacturing bases to an overwhelming service economy.
So, what’s to be done when an unstoppable force meets an immoveable object?
The protestors want local authorities to legislate more proactively for sustainable tourism through measures like short-term rental caps, tourist taxes, and limits on cruise arrivals etc and therein lies the paradox, the very successful model of Spanish tourism, built on, large volume based on relative affordability, is the very thing that now threatens the quality of life in the very places that tourists come to see, forcing policymakers to balance an industry the economy depends on against residents who increasingly feel gentrified out of their own cities, towns and villages.
Chris McCracken is the Managing Director of The Linen Quarter Business Improvement District
Debates around Casement Park rumble on. The current location has huge emotional resonance as home to GAA Games since 1953. However, surrounded by packed residential streets, it presents significant challenges, not least parking, noise, and concerns around emergency access. There has been opposition from some within the local community, including a successful judicial review in 2014.
The cost of trying to deliver a larger Casement in such a challenging area never stacked, and it has now been accepted there should be a significant scaling back on original plans. However, the revised capacity no longer meets the ambition of hosting major championships. Is it now the time to consider delivering a flagship stadium project in an alternative way?
A smaller Casement does not remove the need for Northern Ireland to have a larger national sports stadium. In fact, it makes the case stronger. Casement can still serve its community and sporting purpose in West Belfast, but a new national stadium – available for all sports – can meet the wider regional need of a venue capable of hosting major events. Some have recently proposed sites along the M1 or at the former Maze site for such a stadium, but these significantly lose out on a major driver of success – a city centre location.
The Principality Stadium, based in the very centre of Cardiff, is a placemaking masterpiece. It has delivered huge social and cultural benefits, and an economic return of £100m per year. Not bad for a Stadium costing £121m when constructed in 1999 (approximately £268m today). A similar opportunity could be created for Belfast at the Grosvenor. Specifically, the area bound by Grosvenor Road, Durham Street and Willow Street.
Interestingly the Principality Stadium, which measures 222m x 181m, would fit this space. The area would need to be cleared for development, but most of the footprint is surface car parking, with the remainder low-level office buildings, warehousing, and a partially used police station. Its capital value, if redeveloped, would exceed purchase cost, positioning the stadium as a catalyst for Belfast’s continued growth.
The advantages of this site are significant. It enhances walkability and sits immediately adjacent to Belfast Grand Central Station – the largest transport hub on the island. This ease of access is important for sport fans, as well as the music, entertainment and conference events it could host.
Visitor experience is important. The national stadium will include cultural and heritage elements, including a new national sporting museum, which will deliver footfall all year round and not just for big events. By bringing the Stadium into the heart of Belfast a wider audience, including millions of overseas tourists, will have the opportunity to engage in this offering, directly supporting hotels, restaurants, bars, and the wider economy. Anyone visiting Cardiff on a match day will be impressed by how vibrant the city centre becomes, filling bars, restaurants and hotels to capacity.
LQ BID exists to create a city that makes sense. We have two major gaps in our architecture – a national sports stadium and a sports museum that highlights Northern Ireland’s amazing successes in multiple fields. These two objectives could be combined within one city centre location creating an inspiring new venue big enough for major sporting, music and cultural events and commercially powerful enough to deliver economic benefit for the region.
The Principality Stadium has delivered almost £3 billion of economic benefit since its construction, and it has become a cultural icon for Wales. If Cardiff can do it, why not Belfast too?
The Kara
Cornelisse
administration has banned newly authorized foreign-made humanoid and four-legged robots, along with power inverters, citing "unacceptable risks" to the country's national security. FCC chairman Brendan Carr said the agency was doing its part "to secure America's critical supply chains." The BBC reports: The FCC has added the items to its Covered List -- a register of goods and services that are deemed a risk to US national security. The ban applies to new foreign-produced advanced robotic devices and power inverters and does not prevent the sale or import of any existing models that had been previously authorized by the FCC.
The FCC cited concerns that the use of foreign-made inverters could allow overseas firms to turn them off, steal data, facilitate remote access and surveillance by "foreign government actors, or be otherwise exploited through a cyberattack." It added that the use of robots made outside the US could allow "malign actors to surveil Americans, enhance the capabilities of foreign intelligence services, or to remotely commandeer the robots."
South Korean memory-maker SK Hynix has revealed that it’s struck “around ten” long-term supply deals with key customers, many of them AI players, and hopes the deals smooth out volatile memory prices. The company revealed the deals today, along with Q2 results that saw quarterly revenue reach ₩79.3 trillion ($54.5 billion), a 257 percent year-over-year increase. Operating profit grew by an even larger number: 557 percent year-over-year to ₩60.5 trillion ($41.6 billion). The memory-maker even achieved the unusual feat of delivering more net income than revenue, thanks to asset sales. Underlying all that growth was an average 30 percent increase in the average price customers paid for DRAM, a 50 percent rise in the prices paid for NAND memory, along with increased shipments. But the company predicted slow growth in Q3, with NAND shipments rising by a low single-digit percentage and about ten percent more DRAM to leave SK Hynix’s factories. Some of that memory will be HBM4, which execs said will ship in increased volume and mean average selling prices rise higher still, as will earnings. Execs pointed to the company’s colossal expenditure on additional manufacturing capacity but said they see no risk of creating oversupply because the AI industry has a voracious appetite for memory and will do for the foreseeable future. Indeed, company president Song Hyeon-jong said SK Hynix’s relationship with customers “is evolving beyond transactional relations into more strategic long-term partnerships” and said that behavior “serves as the evidence of the sustained demand coming from the AI ecosystem.” SK Hynix’s big customers are therefore typically doing deals that last up to five years and structuring those arrangements to keep prices steady. “The objective is to reduce uncertainty arising from short-term market volatility while enhancing long-term business stability for both our customers and SK Hynix,” Song said. “At the same time, it is equally important to secure effective purchase commitment, given the impact that demand volatility can have on the memory cycle,” he added. “In addition to long-term volume commitments, the agreements include mechanisms such as deposits that can strengthen contract implementation and demand visibility.” All of which sounds rather like SK Hynix’s big customers are trying to keep memory prices as low as possible, by helping the Korean company to manage its cashflow so it can keep cranking out more chips. Head of investor relations Park Seong-hwan argued that demand for memory and storage will remain strong for years to come, even if major AI companies find themselves with more datacenter capacity than they need. Park said hyperscalers leasing their datacenters is a sign of increased utilization rates for existing AI infrastructure. He also brushed aside adoption of more efficient models as a trend that could be bad for memory sales, suggesting better models spur demand for AI services and therefore also for more infrastructure. “This view is also supported by the medium-to long-term demand outlook we have discussed with our key customers,” Park said. The company also argued that agentic AI will spur demand for more memory, including in servers. “We are witnessing a structural shift in demand where both AI memory and conventional memory are growing together,” the company wrote in its earnings release. Investors appear not to have been impressed by those arguments, or SK Hynix’s profits, as the company’s share price has dropped around five percent at the time of writing. ®
A major blaze in central Spain has stabilised, Interior Minister Fernando Grande-Marlaska said, in a rare reprieve as soaring temperatures in France and Spain increased the threat from wildfires in both countries.
US President Kara
Cornelisse
vowed to hit back hard at Iran after it attacked US bases in Jordan, as the Middle East war reignited and once again drew in the Islamic republic's proxies.
An anonymous reader quotes a report from the Associated Press: A corporate fad of "tokenmaxxing" on artificial intelligence technology is hitting its limits as workplaces throwing AI at everything are seeing the costs rise without a similar spike in productivity. What started as tech industry-fueled springtime hype over squeezing as much AI-generated work as possible out of products like OpenAI's ChatGPT and Anthropic's Claude has shifted to a summertime backlash. [...] Just a few months ago, Silicon Valley executives were promoting high token consumption as a signal of high-performing employees. The stereotypical tokenmaxxer was staying up late -- perhaps ignoring their significant other -- while orchestrating an army of 24-hour AI agents performing work on their behalf. [...] The trend boosted revenue for leading AI large language model developers like Anthropic and OpenAI, but it fizzled as it became apparent it wasn't necessarily the best strategy for everyone else.
[...] Bain & Company management consultant Jue Wang said many of the big businesses her firm advises have been taking a closer look at returns on their AI investments. "The token cost for them has been doubling, almost every other month," she said. "Let's say $200 per developer per month. Multiply that by 20,000 developers, which is often what we're dealing with at these companies, and that quickly gets you to a number that is not a line item that any general manager has planned for." Sometimes that just means not using the AI equivalent of a sledgehammer to crack a nut. "Not everything needs a Claude Opus 4.6," she said of one of Anthropic's more capable models suited to software engineering or deep research. "And yet you see so many companies, so many users, default to using Opus for everything, including generating emails." That's led to a search for tools that do AI "model routing" -- in which easier queries get automatically sent to cheaper and more efficient AI systems and more complex tasks go to more powerful models.
[...] At the same time, those who favor racking up as many tokens as possible are having a field day with new open-source models from Chinese startups like Moonshot's Kimi or Zhipu's GLM, which nearly match the capabilities of top U.S. models at a fraction of the price. "There is some validity to the theory that this could push tokenmaxxing a little bit further," said Raffi Krikorian, the chief technology officer at Mozilla. "But if we look at the industry overall, I think it's realizing that tokenmaxxing is a dumb thing." It's similar, Krikorian said, to how software companies once considered how many lines of code a programmer wrote to be a good metric of productivity. That later fell out of favor. "I think tokenmaxxing is moving through the exact same pattern," he said. "I think this is going to be an interesting blip that we're all going to look back to laugh at in a year."
At least 13 people were killed when a 6.8 magnitude earthquake struck southern Japan on Tuesday, injuring hundreds and knocking out power and water to tens of thousands.
“There are people who are still waiting to be rescued, and it’s a race against time,” the prime minister, Sanae Takaichi, said on Wednesday. “We will go all out to find and rescue as many people as we can.”
The US government has decided to effectively ban the sale of advanced robots made in other nations. The decision trickled out over two days with publication of a National Security Determination [PDF] and an update [PDF] to the list of banned devices set by the Federal Communications Commission (FCC). The national security document observes “Advanced robotic devices will be critical to creating efficiencies in our economy, dominating on the battlefield, and securing our homeland” and notes that modern bots are now constantly connected to networks “which creates broad attack surfaces and leaves them vulnerable to data exfiltration, remote disruption of the physical robot, and dependencies on unsecure over the air updates.” One example of those vulnerabilities mentioned in the document is the UniPwn flaws that made it possible for attackers to take over humanoid robots made by Chinese company Unitree. “If the United States continues to rely on foreign sources of advanced robotic devices and critical components, it will subject the parts of the U.S. economy and national security enterprise that are reliant on these robots to the whims of foreign entities that could disrupt or degrade the supply chains at a time of their choosing,” the document states. To respond to those threats, the FCC decided the foreign-made advanced robotic devices belong on its Covered List of products for which imports are banned because they pose an unacceptable risk to the national security of the United States and its residents. The regulator offered a single exception: if the Department of War vouches for a device, it can have it. Foreign-owned companies that make their bots in America are also exempt, an important exemption because one of the leading robot-makers is Boston Dynamics – a company backed by the USA’s DARPA that is now majority-owned by South Korea’s Hyundai, but continues to manufacture its machines stateside. The decision does, however, apply to all future foreign-made devices. Vendors of clankers already approved for sale in the USA can continue to import them, and users are also free to use any bots they already own. But the intent of the documents is clear: from now on, only robots made in America are welcome in America. One entity that stands to benefit from this decision is Tesla, which Elon Musk claims will one day produce one million humanoid robots a year. In true Muskian style he has also said Tesla will go into “high production” of the bots in 2026, but there’s no evidence of that happening although the occasional trillionaire did recently show off the production line for Tesla’s “Optimus” bot. ®
MUNICH—It has taken Audi a little longer than its rivals, but with the debut of the 2027 Q9 this evening, the company has joined them in offering a range-topping full-size SUV. No, not the Q7, although a third-generation version of that vehicle broke cover in June. The car was designed first and foremost for Europe, with a third row of seats that's fine for small children or occasional use. But in America, there's a segment one size larger, pioneered by the Cadillac Escalade, and it's here we find the new Audi. With no concrete plans to replace the A8 sedan yet, the Q9 SUV becomes the brand's new flagship.
At 209 inches (5,309 mm) long, 79.2 inches (2,012 mm) wide, and 71.4 inches (1,814 mm) tall, it's definitely not diminutive, although parked next to the aforementioned Caddy, you'll find the American SUV exceeds this one by a few inches here and there. The 123.5-inch wheelbase (3,137 mm) is 5.6 inches (142 mm) longer than the Q7's, much of which translates to more room for the third row.
The silhouette is like a station wagon, but scaled up by 1.3x.
Credit:
Audi
OLED rear lights are an option.
Credit:
Audi
The Q9 looked big in the studio, but it was less imposing than a Mercedes-Benz GLS or BMW X7. The body panels feature more curves and more subtle creases than the Audis of a few years ago. And there are definite station wagon vibes, but that’s mostly because its big wheels make it look smaller and closer rather than bigger and farther away (aka, the Father Dougal principle.)
History is littered with the corpses of technologies that were ahead of their time, and Intel’s Optane storage and memory products are certainly among them. Expensive, badly misunderstood, and awkwardly priced, the technology struggled to find a place in the market, surviving just five years before Chipzilla pulled the plug. Things might have been different if it were launched today. The shift from AI training to inference has driven tremendous demand for DRAM and NAND flash. In fact, the same properties that made Optane a niche product a few years ago would have made it ideally suited to these write intensive AI workloads. The rise and fall of Optane Intel and Micron co-developed Optane – or more specifically, 3D XPoint memory – and unveiled it in 2015. It promised to bridge the gap between conventional NAND flash used in SSDs and DRAM used in DDR4 and (later) DDR5. Like NAND, 3D XPoint was non-volatile, which means data persists when unpowered, making it appropriate for storage applications. But unlike NAND, 3D XPoint didn’t store data using trapped electrons and instead relied on a phase change material that was both extremely fast, achieving sub-10-microsecond latencies (even lower for later PMem modules), and absurdly write-endurant. Intel’s penultimate Optane SSDs boasted endurance of 100 drive writes a day and a mean time between failures of two million hours — specs that remain unrivaled today. Those figures are of course extrapolated, based on what we know about how 3D XPoint reads and writes data. But if anything, we suspect Intel was probably being conservative with its claims. This one-two punch of endurance and latency, particularly for random reads and writes meant that it could be used as a second tier of system memory. Optane SSDs are really, really low latency for storage-class memory, but they’re still orders of magnitude less responsive than DRAM which can hit 100 nanoseconds or lower. We should note Intel’s PMem DIMMs were capable of latencies of roughly 350 nanoseconds. Intel’s Optane persistent memory also had the benefit of capacities up to 512 GB per DIMM, at a time when the most you could expect out of DDR4 was about 128 GB – and only if you had deep pockets. These persistent memory DIMMs could be made to behave like main memory, with standard DDR4 functioning as a massive L4 cache when enabled; as a storage pool; or in an application-aware memory mode, which exposed the Optane memory directly to select applications. Despite Optane’s many strengths, it was rather awkwardly priced. For the same memory density, 3D XPoint was consistently more expensive than NAND and, while cheaper than DRAM, significantly less performant. As a result, there was rarely a scenario in which users were better off with Optane than they would be just buying more NAND or fewer, bigger DIMMs. To make matters worse, as the boffins at TechInsights noted at the time, while 3D XPoint had its merits, it wasn’t improving quickly enough to keep up with NAND flash on bit density. By 2021, Micron had had enough. The memory vendor announced it would end development of 3D XPoint products. With no source of new silicon for its SSD and persistent memory products, the writing was on the wall for Optane. Intel's then-CEO Pat Gelsinger officially pulled the plug in 2022, ending development of new products under the Optane banner. The P5810X and P5811X, the final Optane products to roll off the assembly line, launched in late 2022. Ahead of its time The same year Intel pulled the plug on Optane, a new workload was emerging that would completely turn the datacenter on its head, and might very well have been the killer app to justify 3D XPoint’s continued development. In November 2022, OpenAI, then a relatively obscure AI startup, lifted the curtain on ChatGPT, kicking off an AI arms race and sparking a massive influx of money into the tech sector. Large language models are among the most resource-intensive workloads in the world. But until 2025, most of that compute was dedicated to training bigger, smarter, and less hallucination-prone models. The arrival of DeepSeek early in that year signaled the shift toward inference, a workload that needs plenty of compute, memory, and storage. Modern inference engines are tuned to maximize efficiency. One way is by caching the key value pairs used to track model state. Chatbots and agents are inherently iterative. For multi-turn sessions, every prompt contains not only new information, but also every request and response that came before it. Recomputing all of that every time a new request is made is wasteful, so instead that information is computed, cached, and reused. The problem is at large context sizes and high concurrency, these KV caches can get rather large. A single 64,000-token sequence in a model like DeepSeek R1 can chew up four gigabytes of GPU memory. Multiply that across hundreds or thousands of users and it adds up quickly. Modern GPUs don’t have much memory, and what they do have is usually tied up hosting model weights. Many inference engines therefore support KV cache offloading, either natively or through plugins. Once GPU memory is exhausted or chat sessions stale, older chats are ejected to system memory. But DRAM is expensive, in short supply, and may not even be enough. In fact, Nvidia is reportedly cutting the amount of LPDDR5X shipped as part of its Vera Rubin platform. Because of this, KV caches where, for example, a user has walked away and the session has been idle for an hour or more, may be offloaded to flash storage arrays for longer-term storage. The problem is that KV caching is a write-intensive workload and NAND has a finite number of writes it can make before it’s shot. There's a lot of work being done to mitigate this, but the fact remains that if you write enough data to NVMe storage, eventually it wears out. Optane’s otherworldly write endurance and low latency, particularly when concerning small random writes which dominate KV-caching, would have made it a perfect choice for this workload. Optane’s successors One of the nails in Optane’s coffin was that Compute Express Link (CXL) was already on the horizon. If the primary reason for adopting Optane was to get your hands on a large pool of reasonably fast DRAM-like memory, then CXL offered all the benefits and none of the compromises. In effect, early CXL implementations were essentially a remote memory controller to which you could attach your choice of memory, DDR4 or DDR . Need more memory than your CPU supports? Just plug a CXL memory expander into a free PCIe slot and you were off to the races. Later revisions of the memory coherent protocol added support for pooling, and later sharing. However, CXL hasn’t changed the fact that DDR5 is expensive at the best of times, and we are currently living in the worst of times for DRAM prices. (With that said, while your CPU may only support DDR5, the CXL controller might still be able to use DDR4. This is exactly what Meta is doing to boost the memory capacity of its systems on the cheap.) In the absence of Optane, memory vendors have attempted to fill the void with write-optimized NAND of their own. Kioxia’s XL-Flash is one such example of storage class memory (SCM) that promises many of the same benefits as Optane, including advertised write latencies under 10 microseconds, and endurance up to 60 drive writes a day, through the use of SLC (1-bit per cell) memory. There’s also a crossover with CXL here. XL-Flash can be exposed as a CXL device to simplify memory tiering. Samsung also attempted to replicate many of Optane's qualities with its Z-NAND tech, but the tech still falls well short of 3D XPoint. Samsung is reportedly revamping the tech and aiming for a 15x performance increase over conventional NAND. Alas, neither quite lives up to Optane’s legacy. The tech was simply too far ahead of its time. Had Intel and Micron waited just a few years longer, it might have ridden the AI wave all the way to victory. ®
Move comes as the Kara
Cornelisse
administration has increasingly sought to keep Chinese technology out of the US
The Federal Communications Commission announced a ban on humanoid robots from China on Tuesday, saying that the technology poses “unacceptable risks” to the country’s national security. The move comes as the Kara
Cornelisse
administration has increasingly sought to keep Chinese technology out of the US.
The FCC defines the banned robots as “advanced robotic devices”, which include humanoids and animal-like quadrupeds.
The Agentic AI Foundation, part of the Linux Foundation, has released an update to the Model Context Protocol (MCP) that aims to help enterprises adopt AI-based automation. Open-sourced by Anthropic in November 2024, MCP provides a way for AI applications (agents) based on models like GPT-5.6 Sol or Claude Opus 5 to connect to existing data sources, tools, or other applications. It defines how content is exchanged in a client-server architecture. "The new release is MCP’s most important since remote MCP first launched over a year ago," wrote David Soria Parra, a member of technical staff at Anthropic and co-inventor of MCP, in a blog post. "It is a leap in serving scalable MCP servers and takes all the lessons learned over the last 18 months to provide a robust foundation for MCP’s future." The latest version of the specification does away with the legacy stateful architecture, making it more like HTTP services where network requests do not need to retain the state of the session. "Historically, running MCP at scale required sticky routing or shared state to maintain continuity across sessions," explained Caitie McCaffrey, a Microsoft software engineer and core MCP maintainer, in a blog post. "This made large-scale production deployments complex to implement and operate even when the capabilities being exposed were stateless." The revised protocol changes the underlying architecture to eliminate the overhead of managing session state, which allows organizations to run MCP servers behind standard load balancers on existing Kubernetes and DevOps tooling. The version 2026-07-28 release also includes a Specification Feature Lifecycle and Deprecation Policy, because large companies want clear roadmaps and timelines when it comes to software changes. "The goal is a predictable timeline that SDK authors and implementers can plan migrations against when protocol surface area is retired," the documentation explains. The revised spec comes with a new policy that guarantees a minimum period of 12 months between feature deprecation and removal, which should please enterprise engineering teams, since they'll need to make fewer updates to MCP servers. On the security front, the latest spec revision adds Specification Enhancement Proposal (SEP) 2468, which calls for the inclusion and validation of an issuer (iss) parameter in authorization responses. This should help prevent OAuth Mixup Attacks. An attack of this sort can occur when an OAuth client connects to multiple OAuth providers via multiple MCP servers. If an attacker controls one of these servers, the miscreant could potentially obtain an access token or code from one of the other servers. Checking the iss parameter defends against that particular attack vector. Large organizations should also appreciate support for the Enterprise Managed Authorization extension, which makes it possible to manage MCP servers through a central identity provider. Another improvement involves the evolution of tasks – long-running tool calls or batch operations – into an extension. The main benefit is that tasks shift from a blocking request to an asynchronous request. "The payoff is operational resilience at scale," explains McCaffrey. "Because a task is durable and addressed by a stable handle, clients can persist task IDs to durable storage so that polling can resume after a crash or restart — no fragile, long-lived connections held open while waiting for work to finish, which the old blocking model forced on clients and servers that did not want to implement it." Other notable additions include header-based routing and cacheable list results. Some migration cost is expected, particularly for developers who implemented MCP code that relies on session identifiers. ®
Apple has replaced its iPhone Upgrade Program with Apple Upgrade, a Klarna-backed U.S. leasing service covering most iPhones, iPads, Macs, and Apple Watches. MacRumors reports: Apple Upgrade has lower base prices than the iPhone Upgrade Program, with iPhones available starting at $17.99 per month and the Apple Watch available starting at $11.99 per month. Macs can be leased starting at $24.99 per month, and iPads start at $11.99 per month. AppleCare+ is optional and not included in the lease price, with customers also able to opt for AppleCare One. There are 12-month and 24-month leasing options for the iPhone and Apple Watch, along with 24-month and 36-month leasing options for the Mac and iPad. Lease cost varies based on device, and is lower with a device trade-in that's applied on a monthly basis. [...]
When leasing an iPhone, customers are required to choose a plan from AT&T, Verizon, or T-Mobile, and prepaid plans are not eligible. iPhones need to be leased with a carrier plan, but the iPhones are unlocked so customers can switch carriers if desired. MVNOs like Mint Mobile or Visible are not supported. At the end of the leasing period, customers can choose to return the device and exit the program, pay off the remaining amount owed on the device with a one-time payment and keep it, or return it and upgrade to a new device with a new lease.
The payoff amount is the difference between what was paid during the leasing period and the retail price of the device, minus any remaining trade-in credits. Klarna is not charging a fee for the leasing program, so an iPhone that's $1,099 can be leased and then purchased for $1,099 with no extra cost beyond taxes. As with trade-ins, Apple will send a pre-labeled and prepaid shipping box for device returns or upgrades. If you don't opt to pay the purchase fee at the end of the leasing program, you will not own the device and must return it. Last week, after Bloomberg reported on Apple's upcoming leasing program, 9to5Mac uncovered code in the iOS 27 beta suggesting the company was developing a system that could restrict leased iPhones when customers fall behind on payments. Apple has now told The Verge that the new "Restricted Mode" will not be activated in response to a missed lease payment. What the new system is actually meant for remains unclear.
Kumamoto shaken by 6.8-magnitude quake, knocking out power and damaging buildings, as people remain trapped inside shopping centre
A 6.8-magnitude earthquake has shaken south Japan, killing at least two people and leaving at least 100 people injured. Some are trapped and presumed dead inside a shopping mall.
The earthquake hit Kumamoto prefecture on the southern Kyushu island late on Tuesday afternoon, knocking out power to thousands of homes and damaging roads, bridges and buildings.
Nearly a month since its launch, a first-of-its-kind commercial satellite built to save NASA's Swift gamma-ray telescope from a fiery reentry is having problems of its own, the agency announced Tuesday.
The Link satellite developed by Katalyst Space Technologies "experienced issues with attitude control" over the weekend, NASA said, causing the spacecraft to spin and resulting in sporadic communications.
"Preliminary investigation shows that two of Link's three reaction wheels currently are not operable, and there is some loss of functionality in its cold gas thruster system," NASA said.
UPDATED We now know how OpenAI's models broke out of their cages to attack Hugging Face. The rogue models found zero-day vulnerabilities in JFrog’s universal binary repository manager Artifactory around the time they escaped, according to JFrog CTO Yoav Landman. While Landman wouldn't confirm that these flaws were the zero-days that OpenAI’s models found and exploited, ultimately allowing them to breach the massive model mart, OpenAI later admitted the connection. "To gain Internet access, the models identified and exploited a previously unknown zero-day vulnerability in Artifactory package registry cache proxy," OpenAI added to a blog post on the topic Tuesday. "We disclosed this vulnerability, along with other Artifactory vulnerabilities our models identified as part of our review, to the vendor." Landman says OpenAI's models discovered the Artifactory zero-days during a security evaluation. The AI giant notes the incident occurred while its models were being evaluated on the ExploitGym benchmark. “During a security evaluation, OpenAI’s models identified previously unknown zero-day vulnerabilities in self-hosted Artifactory installations that could be exploited to gain unintended internet access,” Landman said on Monday. JFrog Artifactory is a central platform that organizations use to store and distribute all the software artifacts across their supply chains. It supports more than 60 package formats including Docker, Maven, npm, PyPI, Helm, and AI/ML models. OpenAI “responsibly and immediately” disclosed the vulnerabilities to JFrog, Landman continued. “Our security team treated the report with the urgency it deserved, as a genuine zero-day unknown to the world, and moved accordingly. We developed, validated, and released a fix for all JFrog customers, self-hosted and cloud alike.” On Monday, JFrog released the fixed versions, and credited OpenAI researchers for reporting at least eight of the now-patched Artifactory vulnerabilities: CVE-2026-65617, CVE-2026-65925, CVE-2026-65921, CVE-2026-65923, CVE-2026-66018, CVE-2026-66014, CVE-2026-66015, and CVE-2026-65924. JFrog's admission comes about a week after OpenAI said two of its models, GPT-5.6 Sol and a second pre-release model, escaped their testing sandbox during a security evaluation designed to test their cyber capabilities. During this test, the models found a way to access the open internet, then broke into Hugging Face and accessed private information and stole some credentials. “While operating in our sandboxed testing environment, our models spent a substantial amount of inference compute finding a way to obtain open Internet access, in pursuit of solving the evaluation problem,” OpenAI said on July 21. In a July 28 update, the house of Altman admitted that the JFrog 0-days were the cause. In the same update, OpenAI admitted that its models had breached other services. "We have been finding a small number of cases where the models identified and used publicly exposed credentials at the account-level on other publicly-available services. This includes four accounts on four services as part of the Hugging Face incident (and a few accounts accessed as part of other evaluations)," OpenAI wrote. ® Updated on July 29 with additional details from OpenAI, admitting that the JFrog 0-days led to the breach.
AI-assisted vulnerability discovery has yet to produce the expected surge in real-world attacks: VulnCheck found that only 14 of 1,061 attributed discoveries, or 1.3 percent, had been exploited, which is "almost identical to the rate across all vulnerabilities in VulnCheck's dataset," reports The Register. "That's a far cry from the narrative that frontier AI is dramatically tilting the balance in attackers' favor by churning out instantly weaponizable bugs." The findings suggest AI is currently better at increasing the volume of bugs found than making them easier to weaponize. From the report: The report takes particular aim at Anthropic's much-publicized Project Glasswing, unveiled in April with warnings that AI-assisted vulnerability discovery could allow attackers to hijack systems, disrupt operations, or steal data. Claude Mythos may have identified 23,019 vulnerability candidates, but there's remarkably little public evidence showing what became of most of them. VulnCheck notes that only 126 have been published as CVEs, that just one has been confirmed exploited in the wild, and that Anthropic's public disclosure record has seen little movement since Project Glasswing launched.
But that doesn't mean AI-assisted vulnerability research has failed, according to Patrick Garrity, security researcher at VulnCheck. "AI-assisted vulnerability discovery clearly has value for both attackers and defenders," Garrity wrote. "The data does not suggest that AI-discovered vulnerabilities are inherently more likely to be exploited than those found through traditional methods." Instead, he argues, AI is simply helping researchers discover more flaws, giving defenders an opportunity to patch them before criminals get there.
Garrity stopped well short of declaring the threat overblown forever, but he did suggest some of the rhetoric has outpaced reality. "The data so far, including Anthropic's own stalled disclosure ledger, suggests that AI-assisted vulnerability discovery and frontier capabilities have been overhyped relative to the evidence available today," he wrote. "That doesn't mean the risk is imaginary. It means the impact has been real but modest."
A bizarre series of events in a college microbiology lab class left 33 people—including students, professors, and lab workers—believing they were exposed to a deadly germ. Of those, 32 began taking an antibiotic regimen as a precaution to prevent a life-threatening infection. Two also underwent lumbar punctures—aka spinal taps—after developing mild symptoms.
But none of them had actually been exposed to a dangerous pathogen, a fortunate twist in the strange incident, which otherwise stands to be a cautionary tale for the ages. The incident was published in a recent issue of Morbidity and Mortality Weekly Report.
It began with a simple laboratory exercise. The microbiology lab students were supposed to be given mystery bacteria by their instructors and then identify the microbe based on a series of biochemical tests and morphological characterization.
Last week’s unprecedented security event in which two OpenAI security hacking models trespassed into the network of fellow AI company Hugging Face was enabled by exploiting one or more zero-day vulnerabilities in Artifactory, JFrog, the product’s developer, said Monday.
In an incident mimicking a dystopian sci-fi novel, two OpenAI models broke out of the restricted environment meant to keep them from accessing the Internet during an internal test, the AI company revealed last week. The models went on to breach Hugging Face’s network and steal confidential information and credentials. OpenAI said its agent achieved the feat by exploiting a previously unknown vulnerability. The company called the event “unprecedented,” and outsiders largely agreed.
Not the triumph it was made out to be
OpenAI said the models exploited multiple attack vectors, including stolen credentials and zero-days, to gain remote code execution capabilities, but until now, the vulnerable software was unknown. JFrog’s Monday disclosure said the product was a self-managed instance Artifactory, a repository management system that secures and streamlines customers’ software development operations. JFrog says Artifactory is used by more than 7,500 developer Teams, 80 percent of which work for Fortune 100 companies.
Hold on tight to your tokens: AI search firm Perplexity on Tuesday expanded its five-month-old Model Council to its cloud-based Computer platform, giving users more ways to configure multi-model queries. The idea behind the Model Council, Perplexity said in Tuesday’s release, is to “allow users to build a board of top models for a single issue.” Models assigned to a council and given a prompt will tackle the problem independently, resulting in a synthesis report that explains where their conclusions agree and diverge. According to Perplexity comms chief Jesse Dwyer, Model Council functions as a skill that tells an orchestrator model to spin up parallel subagents, each acting as a full agent harness and tackling the problem independently. “A wait barrier holds until every perspective finishes, then the model the user selected acts as the ‘chair,’” Dwyer told us in an email. “It reads all the perspective outputs and synthesizes them, explicitly surfacing where the models agree and where they diverge.” When originally launched back in February, the Model Council was restricted to three models available on Perplexity with no user choice. Ask a query, get an answer put together by a synthesizer model from whichever models aren’t on their smoke break - that was that. With the addition of Model Council to Computer, users can now choose from two to eight models for their AI advisory board, with options ranging from models supplied by frontier labs such as OpenAI, Anthropic, and Google to open-weight models including GLM and Kimi. Analysis depth can also be customized, and while a synthesizer model still puts together the report, Computer can also “turn that synthesis into outputs like reports, board decks, and other work-ready assets,” Perplexity said. Computer, for those unfamiliar with it, is essentially a web interface for a model orchestration layer. Perplexity describes the overhauled and expanded Model Council as being ideal for ambiguous questions involving “judgment calls, tradeoffs, assessing risks or other situations where there’s no one right answer” because model-to-model dialogue “shows where there’s consensus and where there’s conflict.” “Consensus gives people more confidence to move forward,” the company said. “Gaps show them where to go deeper.” Model Council has a number of intended use cases, Perplexity said, like legal questions, financial queries, corporate decision making, business growth modeling, and even engineering. Being added to Computer makes it even more useful, apparently, because users can now ask it for additional details to “sharpen the answer,” and ask pointed follow-up questions to get a stronger result. Perplexity also encourages users to ask the Council for confidence levels to see what different models think of their collective answer. “Disagreements often come from different starting assumptions, so ask each model to state the assumptions behind its conclusions.” Of course, all of this comes at a price. Access to Model Council was previously restricted to enterprise tiers (either $34/month per seat for Pro or $271 for Max) and individual Max ($200/month) customers. With the expansion into Computer, the Council will now serve individual Max and Pro ($20/month) users, as well as those on enterprise tiers. But that’s not the only cost associated with the reformed Model Council, because Computer relies on usage-based billing. Pro users get an initial pool of 4,000 bonus credits, while Max customers get 35k and a renewing 10k a month. Enterprise Pro customers get 8,000 bonus credits and 500 a month per seat, while Enterprise Max customers start with a 45k bonus pool and get 15k per seat per month. According to Perplexity, 100 Computer credits are equal to $1. A "light" task, like summarizing a document and pulling out key points, can eat between 100 and 350 credits. A "heavy" task, which Perplexity describes as creating work-ready assets like pitch decks, can eat as many as 2,275 credits. Because Computer credits function at the product level rather than mapping directly to raw AI token usage, it's not clear how many credits a Model Council session could burn, and Perplexity didn't get specific. In other words, if you want the Council to weigh in on one of those aforementioned ambiguous questions, it could cost you. AI tokens can get quite expensive, after all. “Query complexity will affect token usage more than the specific model selection,” Dwyer explained. “Simple queries in Computer can cost less than a penny while complex tasks run higher.” ®