The Register

What do the first general-purpose programmable electronic computer and greeting cards that play musical jingles have in common? DARPA cites both of them as the inspiration for a project seeking low-resource computing paradigms to be used on future battlefields. DARPA on Friday put out a request for information (RFI) for new low-resource computing (LRC) paradigms and processes that haven't been utilized yet in microscale systems not unlike the tiny chip-and-battery combos found in greeting cards and children's books. Apparently, thinking about those and the 80th anniversary of the ENIAC (Electronic Numerical Integrator and Computer) has DARPA wondering why more useful things can't be made in a tiny form factor. "The physical and financial cost of computation has plummeted so far that it is routinely embedded in disposable novelties," the RFI said, pointing to the fact that a greeting card chip has processing speed and memory capacity vastly exceeding those of ENIAC. "The computation itself has become virtually free; the physical resources required to sustain, house, and power it have become the critical bottleneck." DARPA isn't trying to miniaturize a datacenter, in other words. This is specifically about what it said is a "resource paradox" at the low end of the computing spectrum, and it's one the Pentagon's research arm thinks is ripe for exploitation to power battlefield computing needs. To that end, DARPA said it's seeking concepts that address areas like operating with little power and memory, tolerating unreliable components, and requiring little technological sophistication. Regarding the latter, DARPA said it wants systems that can be built using low-precision manufacturing techniques, legacy fabrication processes, and/or "primitive technological ecosystems." Responses need to address at least one of those issues, DARPA said, and ideally (though optionally) one of its defined logistical predicaments as well. By those, DARPA is referring to the sorts of things that would constrain operation in environments where the US military typically operates. Concerns include low-trust environments where data sources and system components may not be trustworthy, operating with the minimum necessary privileges to avoid root access concerns, simple user experiences that even the average grunt can understand, and what may be the most complex issue of them all: self-hosting. We're not talking about soldiers in the field managing their own websites here, though. This is actually about autonomous systems. DARPA wants cheap, tiny, and reliable devices that are still "capable of native, user-directed, autonomous self-programming and self-modification without reliance on external cross-compilation toolchains or host machines," as well as "architectures that allow the system to adapt, recompile, or generate its own operational code entirely on device." That's quite the ask for microscale, low-end hardware, but hey – it's not like DARPA hasn't been the proving ground for a bunch of technology we all use on a regular basis. Someday soon even your average electronic greeting card might host a mini AI if this program proves successful. ®

The global memory crisis has developed a new twist as buyers turn to "legacy" products such as DDR2 and DDR3 to meet demand, according to market watcher TrendForce. The Taiwanese firm says DRAM buyers are turning to older products to secure larger supply allocations, driving up prices for components including DDR2 and DDR3. As Reg readers will be well aware by now, the AI craze has led to memory chipmakers prioritizing production of more profitable HBM and server DRAM silicon to power AI infrastructure, leaving a shortage of the mainstream memory types needed for PCs, smartphones, and other devices. As a result, prices have risen for DDR4 and DDR5 modules – if you can even find them – resulting in hikes in the cost of kit such as PCs, which are up by double figures, according to some estimates. Continued shortages of everyday DRAM components and rapidly rising contract prices have prompted some hardware makers to downgrade memory specifications to control system costs, TrendForce claims. In some cases, DDR4 designs are being replaced with DDR3 solutions, while certain DDR3-based products are being redesigned to use DDR2. We find it hard to believe that PC makers would ship systems with memory types so old or that modern processors would support them, so it is likely this applies to other kinds of device. Now the market intelligence operation estimates that DDR2 contract prices will rise by approximately 55 to 60 percent for the second quarter of 2026, followed by a further 35 to 40 percent increase in the third quarter. This is happening because customers are desperate to secure more reliable supplies, adopting lower capacity configurations or turning to older memory generations. Consequently, the supply shortages are now rippling through the memory market and starting to affect even legacy DRAM products. Key suppliers of DDR2 components include Winbond and Elite Semiconductor Microelectronics Technology (ESMT), based in TrendForce's home turf of Taiwan. However, Winbond is gradually winding down DDR2 production and reallocating capacity toward more high-margin products such as DDR3, DDR4, and LPDDR4, it says. But ESMT plans to maximize DDR2 production within its existing allocation at wafer maker PSMC. The firm is understood to be concentrating resources on this segment to enhance profitability and help offset the supply gap created by Winbond’s withdrawal from the DDR2 market. Some of the big memory makers are planning to increase capacity, but only slowly. Korean giant SK hynix aims to double silicon wafer output capacity over the next five years, while US biz Micron expects "meaningful new capacity" at its new Virginia fabrication plant in 2027 and 2028. ®

Google Cloud Summit came to London last week, and we took the opportunity to sit down with database execs Sailesh Krishnamurthy (VP engineering) and Yasmeen Ahmad (product executive Agentic Data Cloud). The event was wall-to-wall agentic AI, and true to the theme, Ahmad told us that "we're putting agents at the center ... with the goal that humans are not going to be using data platforms in the next three to five years. It’s going to be humans orchestrating agents, and agents actually doing the work." One of the key AI-driven changes, Krishnamurthy said, is that when retrieving data "it’s not so much about getting the exact results, but getting the best results." For developers skilled in crafting SQL queries that get precise results in the most efficient way, the notion of inexact queries that go through some sort of non-deterministic and compute-expensive parsing may seem like a step backwards. "If you have exact questions, you need to be able to provide exact answers," Krishnamurthy told us. "But I think inexact questions are what people are also going to expect. When you think about agentic workloads and operational databases, you want to be able to ask more flexible questions." An example might be a natural language query that takes into account context, such as previous interactions. Krishnamurthy described "AI native infrastructure," including vector indexing, text indexing, and graph technology where "you combine structured and unstructured data, you have to be operating in terms of inexact results and data quality." The company is also investing in the "knowledge catalog," formerly called Dataplex, which is enterprise search now also treated as context for LLMs (large language models). Knowledge catalog aggregates organization data across multiple sources including structured and unstructured sources. Krishnamurthy said that exact SQL queries are not going away, and that sometimes a "fuzzy question in natural language" might generate an SQL query with exact results. How do you verify that AI-generated SQL is producing the results you want? “The answer is the same, not just about SQL, but about many AI-related things," said Krishnamurthy. “The answer is a set of evals you have to maintain ... you might start with something where some results work well and some don’t. And then you have to keep iterating on your blueprints and other pieces of context until your eval set is 100 percent working well." By eval set, Krishnamurthy means "a set of questions that are representative tests that users may have, and what is the right query that is generated associated with it, and then a determination of is this query, is this answer correct or not?" Google SQL as used in its distributed Spanner database, PostgreSQL-compatible AlloyDB, and in the BigQuery data warehouse engine now has AI functions such as AI.IF, which evaluates a condition described in natural language and returns true or false. The prompt value is evaluated using a Gemini LLM; and could return an error or null if the model fails such as when unavailable or out of quota. The inefficiency of functions like AI.IF is a problem, but there are possible solutions. One is the idea of proxy models, which Krishnamurthy described as "a tiny model in the database." A proxy model is trained on the fly, based on a small sample of the data. The query engine evaluates the results from the proxy model, and if good enough, uses it for inference in place of a call to the LLM. According to a paper on the subject proxy models "consume about 400x less tokens, and the latency goes down by 30x-100x." We asked Ahmad why she believes humans will soon not interact directly with Google’s data platform. The answer, she said, is based on the idea of intent-driven engineering. "Three years ago everyone was doing prompt training classes. Really, these models were co-pilots or assistants. Now these models are doing multi-step execution, parallel execution, handling complexity. So you can define an intent, a goal, an outcome, and the model will figure out the steps to get there." According to Ahmad, humans will act as orchestrators, thinking about business outcomes, and models will do "the hard graft of figuring out the low-level data wrangling." She said that today’s staff need to be skilled not so much in prompt engineering, but rather using AI for spec-driven development. "The focus for the human is getting to the right plan and iterating with the model on what is the right way to think about the problem." In business intelligence, she said, companies will move away from dashboards because they only "serve the first layer of predictable questions." In their place will be "conversational analytics for business users." She believes that unwelcome aspects of generative AI, such as hallucinations and prompt injections, are mitigated by improved context, such as from Knowledge Catalog. "I have customers who have got 90 percent plus accuracy with conversational analytics, but that was not the case 18 months ago when the models would get one out of every two questions wrong because they would not have that context." A problem here is that even over 90 percent accuracy is not good enough if you are, for example, a customer of a company with heavy AI adoption confronted with a blocked transaction or other rejection because of an inaccurate response. Another issue is that injecting AI into every interaction means paying for tokens on top of the base compute and storage resources traditionally consumed by cloud database platforms. Higher productivity and reduced staff costs may more than compensate, but this cannot be taken for granted, particularly as reducing the skill barrier with features like conversational analytics also tends to increase usage. Giant cloud providers like Google though have plenty to gain. AI, Krishamurthy told us, is driving growth in data storage as well as token usage. He described "a huge overall growth in the business because everyone needs data … Anthropic, for example, rely on BigTable to store all their prompt information. They have other workloads too which are not public." Two metrics he is permitted to talk to us about, he said, are that Spanner "now runs 7 ½ billion queries per second at the peak … a year back Spanner might have been 5 billion queries per second." Spanner, he said, "has about 23 exabytes of data. It’s the same with BigTable, roughly 7 billion queries per second and double-digit exabytes." Models make more queries, he said. "Instead of taking the user request and just sending one query, one pattern I’ve seen is a model will send five different queries … it’s hard to say exactly what is happening because the models are trying different things."®

A Canadian power utility says customer data may have walked out the door during a security incident, but isn't yet saying whether the intruders got anywhere near the systems responsible for keeping the lights on. London Hydro, which distributes electricity to more than 160,000 customers in and around London, Ontario, said on Saturday that it is investigating a data security incident that "may have impacted a portion of personal information on some accounts" and has started notifying affected customers. The utility said the potentially exposed information includes names, addresses, email addresses, phone numbers, account and billing numbers, service addresses, pricing plans, contract start dates, and meter information. The good news, according to London Hydro, is that the incident did not involve banking information, payment card details, dates of birth, government-issued identification numbers, or other sensitive financial data. The less good news is that the company has disclosed little else. Its statement focuses on customer information and contains no indication that operational technology or grid systems were affected. London Hydro has yet to explain what systems were compromised, how the incident occurred, whether data was stolen or merely accessed, or how many customers may have been caught up in the incident. The haul may not include bank details, but it contains enough account information to make a fake utility bill, payment demand, or customer service call look considerably more believable. London Hydro is warning customers to watch for suspicious communications, unexpected bills, unfamiliar account activity, or requests to change payment arrangements. The company also reminded customers that it does not ask for banking details by email, phone, or SMS. The Register asked London Hydro when it discovered the intrusion, whether information was exfiltrated, how many customers were affected, whether ransomware or extortion was involved, whether any third-party systems were implicated, and whether operational or grid-related systems were touched during the incident. At the time of writing, London Hydro had not responded. The company has drawn a fairly clear boundary around the customer information that may have been exposed. Where the attackers went and what else they may have touched remains unclear. ®

Microsoft has issued a stark warning for Windows users everywhere: "Get ready for Windows 11, version 26H2." Unless, of course, you're running Windows 11 26H1, in which case 26H2 will not be offered and you'll have to wait for a later upgrade path. 26H2, due for general availability in the coming months, uses the same Windows core as Windows 11 25H2 and Windows 24H2. As such, it will be delivered via an enablement package that switches on the new features and updates the versioning. 26H1 is an altogether different beast, aimed at devices based on Qualcomm's Snapdragon X2 Series processors and other new hardware platforms. Other than the version number change, users will be hard-pressed to notice much difference between 25H2 and 26H2, which, frankly, is no bad thing. A bit of stability wouldn't be a problem for Windows users, even if the monthly updates, where most new features are delivered these days, can be somewhat hit-and-miss in terms of quality and reliability. Windows Insiders can take a look at the new release now – the brave souls in the Experimental channel will see their versioning updated to 26H2, and there are also paths from the Dev and Beta channels. Also lurking in the previews is an update to Windows Search that will bring a sigh of relief to many users. In an Experimental Insider release just over a week ago, Microsoft wrote: "Finding apps is more forgiving. Search is better at handling typos, dropped letters, extra letters, and partial words for apps. Queries like 'utlook' can still find Outlook. "Settings results are improving. We've made ranking improvements to help more relevant settings appear higher in results." All well and good. Search has been infamously awful in Windows for years. However, March Rogers, partner director of product design for Windows at Microsoft, later clarified: "Local files are prioritized in the new search updates. Also, if you want, you can turn off web suggestions entirely." Tweaking Search to ditch web suggestions has been possible for a while, if a user knows where to look, but bringing the setting front and center (according to Rogers, it can be turned off in Settings > Privacy & Security > Search) is a welcome move. Otherwise, 26H2 should be a straightforward release, with little in it to alarm users and administrators – Microsoft saves those for monthly updates. Support for Windows 11 24H2 will end on October 13, 2026, for Home and Pro users (Enterprise and Education users get another year – until October 13, 2027). Microsoft's support lifecycle means 26H2 will get 24 months of support from its release for Home and Pro, and 36 months for Enterprise and Education. ®

Long have we known that technology speeds up everything, and it's no different with British prime ministers. In the last ten years, the UK has gone through six prime ministers; it took 34 years to get through the previous six. Nonetheless, Sir Keir Starmer had hoped to make an impact on the UK tech industry and the government's use of technology during his tenure. Given the results, it may be a mixed blessing that, after nearly two years, he has achieved little. The signs were not good from the start. Desperate to invest without raising taxes, his government claimed it could find £45 billion in efficiency savings by applying AI and automation in the public sector. It even gave the cutesy Humphrey moniker to one AI bot in a reference to the much-loved sitcom Yes Minister. No Chance Minister would have been more accurate, as experts poured scorn on the estimate and, just a few weeks ago, a committee of MPs warned that such hype hinders effective digitization of government services. The botched announcement, which had little impact, set a precedent for how the Starmer government's technology policies would unfold: grand promises, too little thought, and contradictory priorities. Take another high-profile announcement. In July last year, then-technology secretary Peter Kyle signed a pact with Google Cloud with the promise to "upskill" as many as 100,000 civil servants in the latest tech by 2030. The agreement, announced at the Google Cloud Summit London, promised to help meet the government's target of having one in ten public officials designated "tech experts" to implement civil service reform. Google DeepMind would also work with technical experts in government to support them in "deploying and diffusing" emerging technologies. But the government later clarified that its deal with the Chocolate Factory "doesn't comprise any specific commercial agreement" and there was no payment schedule from the government. Google's apparent deal was just one tied to a $42 billion (£31 billion) trade pact announced to coincide with a visit from US President Trump. Just a year later, the government now wants to focus £400 billion in public spending power on UK sovereignty and innovation, including investment in tech. Nonetheless, the government has shown repeatedly that it is unable to hold its largest tech suppliers to account, as the supplier at the heart of the historic Horizon scandal – Fujitsu – has yet to contribute to compensation for victims of one of the nation's worst miscarriages of justice, one that, in fairness, predates Starmer's time in office by more than a decade. Also demonstrating the government's tendency to promote conflicting priorities is its so-called pro-growth approach to regulation. Yet, in March this year, the chair of the Competition and Markets Authority's cloud inquiry quit, citing the slow pace of implementing recommendations outlined in a report it published in 2025. As if to demonstrate its inability to resist the temptation to introduce more regulation when it gets a good headline, the government has set out plans to ban under-16s from social media platforms. In 2024, the finance minister promised the NHS £3.4 billion in IT investment, claiming it would unlock £35 billion in efficiency savings by the end of the decade. But the investment is under scrutiny from His Majesty's Treasury and requires an act of Parliament to remove the role of NHS England in implementing health tech policy. Starmer will perhaps be best known in tech circles for resurrecting the idea of digital ID cards, a classic in the solution-looking-for-a-problem genre. In September last year, the Starmer-backed policy promised that all legal residents would use digital identity to prove their right to work by the end of the current Parliament, which could run until August 2029. But by May this year, the plan, which leaves questions over its funding unanswered, was slammed by MPs as a poorly planned "fiasco." In the next few weeks, or possibly days, the UK will get some clarity on the identity of the next prime minister. Front-runner Andy Burnham has already spoken out against digital ID cards, suggesting plans might be nixed if he wins power. Nonetheless, UK citizens can look forward to another ride on the government tech hype cycle. The next prime minister will be caught in the trap between unwanted tax rises and much-needed public investment, just like the last. A politician in a tight spot seldom keeps their hands out of the cookie jar of tech promises for long. ®

The Brazilian National Secretariat for Civil Protection and Defense (SEDEC) and Federal Police (PF) are investigating a suspected hack of the country's emergency alert system after an unauthorized "extreme" alert pinged devices across the country. Defesa Civil Nacional confirmed that its dispatch platform, often used to inform the public about severe weather events, was taken offline in the early hours of Saturday, June 20, after Brazilians reported the alert, which read: "Alerta extremo - Defesa Civil:misantropi4." "Misantropia" is Portuguese for misanthropy, hatred of humankind. The message reached an unknown number of devices, with reports coming from São Paulo, Rio de Janeiro, Paraná, and the Federal District. Civil defense authorities in all four areas confirmed the alerts were bogus and likely stemmed from an attack on the system overseen by Brazil's National Telecommunications Agency, Anatel. "Anatel clarifies that the alert messages received by mobile phone users during the early hours of this Saturday were not issued by the competent authorities responsible for the population alert system," it said in a statement. "There is currently no reason for concern on the part of the population as a result of the messages received." The agency went on to say that it remained confident in the alert broadcast system's capability to help save lives during periods of disaster. National Civil Defense officials have not confirmed whether anyone responsible for the suspected breach has been identified or apprehended, although they are not believed to be part of government staff. "The Defesa Civil Alerta dispatch platform was taken offline at 01:30 this Saturday (6/20), after suffering a breach and issuing an alert to various regions of the country, remotely ordered by someone outside the National System of Protection and Civil Defense," the department stated. "The message issued was of the Extreme Alert type and contained the word 'misanthropy' – which means hatred of humanity. It is likely a hacker attack." A Defesa Civil Nacional chief said in a press conference that a new dispatch system was already under development that would place greater emphasis on security and preventing unauthorized intrusions. The agency also committed to relaunching the affected system as soon as possible after ensuring it is properly secured. ®

Microsoft has managed to break a very basic Outlook function – the ability to include the previous email in a reply. The issue, which afflicts Outlook for Mac, was introduced in version 16.110, build 26061317, which rolled out to users last week. Before the update, the original message appeared in the reply window. After the update, nothing. One affected user said: "It makes it impossible to have a proper email thread because the recipient can't see the conversation history." Another remarked: "This is an extremely frustrating bug in this last update, and I hope a fix is on the way. This is frankly quite disappointing that something major like this completely evaded the software testing process before it was being rolled out." The solution, according to one of Microsoft's forum moderators, is to roll back to a previous version and turn off auto-updating until a fix is released. Great for users in full control of their devices – not so good for anyone with a managed device. Administrators with fleets of Macs running Outlook should brace for helpdesk tickets. In some instances, having a user copy and paste the salient bits of the email they are responding to might not be such a bad thing. We've all had emails that required epic amounts of scrolling to find what started the conversation, so forcing users to think about what they actually need to include is no bad thing. However, disrupting user workflows without warning – well, that is undoubtedly a bad thing. This is, after all, one of the most basic things an email client needs to do, so shipping a product with a bug that breaks this functionality says more about Microsoft's approach to quality than anything else. The Register asked Microsoft for its take on the matter, but the tech behemoth has yet to respond. Microsoft has already delivered a kicking to Apple hardware users in recent weeks. Office 2019 will cease to be useful on the platform in July, or enter a "reduced functionality mode" in Microspeak, with the only option available to users keen to remain Office customers being an upgrade. However, as this latest bug demonstrates, running Microsoft's latest and greatest carries its own risks. ®

A Canadian healthcare organization has apologized after its IT team carried out a phishing test falsely offering staff an additional paid day off work. Newfoundland and Labrador Health Services said the phishing test was sent to employees and physicians, acknowledging the theme was inappropriate. "We acknowledge the approach taken in this particular exercise was not appropriate, and we sincerely apologize to employees, physicians, and union representatives," said Ron Johnson, interim CEO at NL Health Services. "We value the feedback and are reviewing how future awareness exercises are developed and communicated. It is important they reflect employee and physician perspectives, as well as our organizational values, to foster a respectful and supportive workplace culture." The test came during an already fractious period for healthcare staff, who had recently worked long hours to launch the new software system CorCare across the organization. NL Health Services referenced CorCare in the test email, thanking staff for their hard work on the launch. The email contained a button to click to redeem an additional paid vacation day, but clicking the button resulted in a fail mark. The Registered Nurses Union (RNU) in Newfoundland and Labrador said the test was especially insensitive since nurses and other healthcare professionals were already struggling to secure paid time off. Burnout and staffing shortages are rife in the healthcare sector – two factors referenced by RNU president Yvette Coffey in her response to the news. "Yes, we have heard concerns from members about this, and frankly, I understand why they are upset," she said. "Nurses and other healthcare professionals have worked through enormous pressure over the last number of years, including ongoing staffing shortages, burnout, organizational restructuring, and the challenges connected to the rollout of CorCare. To use the promise of an additional paid day off as the hook for a phishing exercise was in very poor taste." Coffey added: "Cybersecurity education is important, but it needs to be done with judgment and respect. There are many ways to test phishing awareness without exploiting the very real stress, fatigue, and frustration healthcare workers are experiencing." Johnson told reporters at a press conference that the test "missed a mark," and promised to investigate how it was allowed to be sent. "What happened here, obviously, is that all the lenses that were required to review the scenario weren't placed on it," he said. "It's not reflective of how we value our employees." With cybersecurity awareness being incredibly important in critical infrastructure organizations, some IT experts would argue that these kinds of tests are valuable. Cyberattacks on hospitals and healthcare facilities can lead to devastating consequences, including vital procedures being canceled, service downtime, and in the rarest cases, death. However, as others have previously pointed out, there isn't much evidence linking fire-drill-style tests to improvements in organizational security. ®

INTERVIEW On August 5, a spent Falcon 9 upper stage is expected to crash into the Moon. Humanity has not done a great job of looking after the space around Earth. What could be done differently around the Moon? How about lunar scrapyards? Chiara Manfletti, CEO of satellite collision avoidance business Neuraspace, reckons one option could be to pick specific locations on the Moon's surface as dumping grounds where material can be deposited, collected, and recycled. "I think it's actually more sustainable than, for example, re-entry into Earth's atmosphere. Because, in the end, they are resources that we could reuse." Manfletti is looking into the future, but in a few decades – if the plans of space agencies and their commercial partners are realized – the space around the Moon could rapidly become crowded. Neuraspace is all too aware of the situation around Earth, making the CEO well-placed to comment on what has and hasn't worked well in Earth's orbit and how things could be done differently around the Moon. Around Earth, some orbits are protected and some are not, "which I think was never a good approach," says Manfletti, "because with the amount of objects going up, each and every orbit should be seen as protected." There are graveyard orbits, where spacecraft can be placed as an alternative to a controlled re-entry, although Manfletti says the approach is not particularly sustainable. "Maybe one day we can reuse the material there." The problem is how to persuade governments and industry to work together. "We don't have a single body," says Manfletti, "that can speak, and then everybody says: 'Oh, fantastic! They have spoken, and we shall do it!' It's not how it works." When we spoke to Manfletti in 2024 about the state of the space around the Earth, she wasn't keen on regulations, partly because getting every nation and company to follow the same rules would be nearly impossible. A similar situation applies to the Moon, though there is an opportunity to do things differently. "I think there are soft ways of going about it, where you see strong stakeholders that don't actually have legal power to impose. They can move things in the right direction. "And all of a sudden, there's more of a demand for sustainable activities in orbit. And I think that around the Moon, in a sense, one could go that way. "So who are the bigger players? It's the US, obviously, China… India is also starting to have a strong Moon program, Europe. So it's questionable whether China will put those requirements in, but I could imagine that Europe, the US, and India could. "And that would certainly raise the bar as to what should be done." Neuraspace uses AI and ML models to predict and compute the probabilities of collisions between satellites and other objects in Earth orbit. The number of objects has increased rapidly since we spoke to Manfletti in 2024, but it isn't the increase that took her by surprise. It was the sudden public awareness of space infrastructure. "I think one of the things that perhaps was most surprising was the interfacing between space activities and the Ukraine war, which really raised awareness in terms of the importance of space. "Everybody was talking about the spoofing and jamming of Galileo as von der Leyen was traveling, and the importance of space infrastructure became something that everyone was talking about. "That, for me, was surprising. You always hope that people will understand the importance of space, but it just happened!" As traffic to the Moon is set to increase, there is an opportunity to examine traffic management around Earth and consider what worked, what didn't, and what could be improved this time around. ®

Veteran tech website Gizmodo confirmed a compromise on Saturday after readers reported ClickFix malware prompts appearing on article pages. Users posted screenshots of fake CAPTCHA windows appearing on Gizmodo's site. The attack aims to fool users into running malicious code via their terminals. According to Proofpoint threat researcher Tommy M, the attack was seemingly launched by an affiliate of ErrTraffic, a ClickFix-as-a-service program that allows attackers to deliver whichever malware they choose. He said the ClickFix prompt was tailored to each user's OS. The Windows version attempted to install the NetSupport RAT malware, which abuses the legitimate NetSupport Manager tool to gain access to affected systems. Darktrace says NetSupport RAT can also be used to exfiltrate files from affected systems and to load additional payloads, such as other malware strains and ransomware. The macOS version had a payload configured but appeared to be broken, requiring a password to open a ZIP archive. Gizmodo said the attacks were being displayed only "briefly," and the timeline of user reports, which span just a few hours, suggests that was indeed the case. "We identified and resolved a security incident on our site earlier today," the outlet said. "A compromised account was exploited to inject a malicious script, briefly exposing users to scam content. The site was taken offline immediately, the script removed, and the account secured. "We're back up. If you notice anything unusual, reach out." The Register confirmed that the website is no longer serving ClickFix prompts as of Monday. ®

Ask any parent: banning kids from doing anything doesn't work. On the contrary, it's just about the most powerful way to motivate them. As The Register reported last week, the UK government is planning on kicking under-16s off social media. The move echoes a similar one by the Australian government, which we reported at the end of 2025 – noting at the time that "nobody thinks it'll really work." Perhaps, though, there's more to it than at first sight. Perhaps this is some sinister mandarin's long-term plan to improve digital literacy and technology awareness among British, and indeed Australian, youth, using a well-established form of mind control far more effective than the CIA's MK-ULTRA: reverse psychology. Or, as the late great Sir Terry Pratchett called it, "headology." Although this quotation from Witches Abroad is about witches, he could have been talking about teenagers: "And that ain't easy, with people like them. Got to use headology. Got to make 'em send 'emselves. Tell Esme Weatherwax she's got to go somewhere and she won't go out of contrariness, so tell her she's not to go and she'll run there over broken glass." Given a certain type of teenager, we can think of no more powerful incentive to self-education than to tell them that a powerful adult somewhere doesn't want them to see something fun or potentially naughty. The direct result will be an urge to learn about geolocation, virtual private networks, identity verification, biometrics, and everything to do with accessing bits of the internet that they're not meant to. The legislation specifically says: "The ban will therefore include platforms like Snapchat, TikTok, YouTube, Instagram, Facebook and X." That sounds like a lot of the fun bits to us. Or perhaps not, depending on your perspective. This especially pertains to restrictions that apply to those bits of the internet with lots of, shall we say, flesh tones. It reminds us of the aid package for VPN providers known as the Online Safety Act, which rapidly drove up demand 14 times. It also reminds us of a couple of lines from another Discworld novel, The Colour of Magic. "Rincewind got down on one knee, the better to arrange the picture, and pressed the enchanted lever." "The box said, 'It's no good. I've run out of pink.'" "A hitherto unnoticed door opened in front of his eyes. A small, green and hideously warty humanoid figure leaned out, pointed at a color-encrusted palette in one clawed hand, and screamed at him." "No pink! See?" screeched the homunculus. "No good you going on pressing the lever when there's no pink, is there? If you wanted pink you shouldn't of took all those pictures of young ladies, should you? It's monochrome from now on, friend. All right?" Well, quite. ®

The legal battle between Broadcom and Tesco isn't so much a contract dispute as a serial TV drama. An aggressive organization takes over a renowned operation with blue chip clientele, and puts the squeeze on. Prices go up, customers are encouraged to move to new products if they know what's good for them, and the old ways are shredded. The new mob reckons it has customers locked in. Their tentacles reach deep into the heart of corporate IT infrastructure, and who wants the trauma of a heart transplant when you can still pay for the medicine, right? Lots of little people will move on, even some of the big fish, but who cares. Triple the cost and lose half the customers, and you're still up on the deal. Eat it up. Only, some of the marks refuse to play the game. One is the UK supermarket giant Tesco plc. Citing contractual violations and unfair practices, including the revocation of support options and perpetual licenses, it tells Broadcom it'll see it in court. This is a threat to the authority of the new regime, and backing down is not an option. Broadcom has seen off corporate leviathans like AT&T. A British grocer will do the sums and see sense. The British grocer is not seeing sense. It's seeing red. In the latest episode, new court filings up the ante, with Tesco revealing numerous offers from Broadcom and why they are displeasing, a massive internal push to ditch Broadcom altogether, and the dangers and costs of a forced march to relieve the siege on company-critical systems. Broadcom's bluff is being called. Tesco, it turns out, is not an enemy you take on lightly. It's been around for about 70 years longer than Broadcom, going from a market stall to the UK's biggest supermarket. It makes more money than Broadcom, with revenues of £73 billion ($96 billion) compared to $64 billion (£48 billion). There's selling chips, and then there's selling chips.* These are just numbers. When it comes to contractual shenanigans with suppliers, Tesco has form. Ten years ago, a regulatory investigation found that it had itself been treating suppliers unfairly, delaying payments, making unilateral deductions, and engaging in other improper billing practices. Tesco apologized, paid £1 million ($1.4 million) costs, and restructured its supplier billing. Whether it's the zeal of the reformed or institutional memory of how to play, Tesco is remarkably disinclined to brush Broadcom under the carpet. Sadly, the series finale will be as anti-climactic as Game of Thrones' terminal detumescence. Not for nothing has Tesco estimated the rushed migration will be finished at the end of 2027, which coincides in a non-coincidental way with the start of the actual court hearing. By then, of course, the full cost of the extraordinary measures will be known and Tesco's lawyers will present some truly cosmic estimates for damages. Broadcom will fold, either on the courtroom steps or – if it's sensible – long before. A secret deal will be done. There is always a power imbalance between supplier and client, and this drama shows what happens when one side misjudges the other. With corporate IT infrastructure, bad things happen if the supplier overestimates its importance or if a customer underestimates it. It's bad news indeed if you can't afford to break a supplier lock-in, or if your business can't afford to lose one particular client. IT has been a paragon of this equation, going through cycle after cycle in which power imbalances brought down the big and the small. IBM once owned corporate computing, NetWare owned the server space, Nokia reigned in mobility. The very best time to be aware of history is when a future technology beckons. Virtualization was once the future, decoupling software from the hardware it ran on, enabling efficiency, flexibility, and eventually the revolution of life in the cloud. VMware was a trusted partner in the virtualization revolution, respectful of its clients and respected by them. Technically and corporately trustworthy and responsible, right up to the point a venture capital outfit disguised as a chip company took over and busied itself with weaponizing assets. The joker in this particular game is that while virtualization is fundamental to modern corporate infrastructure, any particular virtualization solution is replaceable. The VMs do the work, and they don't care what's beneath them. It may be painful to realize that, but it can be done. Broadcom's bet is that this pain makes VMware a one-way street; Tesco's bet is that not only is that not true, but the pain can be reflected back to its source. What if a core technology really was a one-way street? What if it involved such a rewiring not just of an organization's underlying tech, but the very nature of the relationship between that org and the tech that powers it? What if it ripped out not just layers of the stack, but the stack itself and the expertise behind it? What if it absorbed so much of every resource that powered the corporate IT market that it choked out the competition? What if all this were true, and then that tech failed? Worse, what if it succeeded? These aren't hypotheticals. This is the game, right now. What Broadcom versus Tesco shows shows us are the real rules, where it can go badly for either side of the supplier/client equation when reasonable assumptions are changed. Like all good dramas, it makes explicit conflicts we recognize but can't make explicit. Big decisions are coming. Enjoy the show. ® *Chips, as in "fries."

WHO, ME? The world of work is weird, so The Register records the worst of it every Monday in a reader-contributed column we call "Who, Me?" in which you admit to mistakes, and reveal your escapes. This week, meet a reader we'll Regomize as "Terry" who told us of a summer job he worked in the 1980s. "It was at a municipal IT facility, and everyone had a specific job to do, and woe unto you if you stepped over the line," he wrote. Terry illustrated the workplace culture by telling The Register that not long before he started work, the facility had just replaced a punch clock with a clipboard on which staff recorded the times they arrived at and left the office. "Everyone wrote 08:00-16:00, but being a young whipper-snapper, I cleverly wrote 07:53-16:01, or 08:02-16:04. That got me hauled off to face the boss for being late." When Terry wasn't tangling with management, his job involved programming an IBM mainframe. One day, he needed to move a program from one drive to another, a task that the mainframe front end didn't support. "A colleague told me there was a MOVE command to do the job, so I ran it and my file disappeared from the source drive as expected," Terry wrote. But the file never reached the destination drive, so Terry's work was lost. "Always one to collect receipts before taking any action, I looked into the MOVE command and found it was a batch program." A very bad batch program with no error checking or safeguards, so Terry rewrote the script so it wouldn't delete data. He sent his changes to the mainframe sysop team, which he assumed would be pleased because his changes prevented data loss. "Instead, I was hauled off to the boss's office again," he told Who, Me? "They wanted to know why I was spending valuable time doing someone else's job. I said I had to because those people weren't doing it right." That answer went down badly, so Terry tried again. "I lost an hour's work and wanted to save my colleagues from the same fate," he said, before again being told to just do his job and take the matter no further. "Lesson learned: just do your job," Terry wrote. "I did my job and got another one elsewhere after that summer was done." Has going above and beyond the call of duty landed you in trouble? If so, it's your duty to click here to send us an email about your experience so we can consider it for a future instalment of Who, Me? ®

KETTLE It's been a week since the Trump administration established a de facto ban on Anthropic's Mythos derivative, Fable 5, and the more that comes out about the move the more it seems like Anthropic employees talking amongst themselves were on to something: Is the government just picking on the company? This week on the Kettle, host Brandon Vigliarolo and Reg cybersecurity editor Jessica Lyons chat about what's going on with Mythos and Fable, what role Amazon may have played in justifying the government's move, how a prominent cybersecurity expert is calling the government's foul, and what this whole thing might mean for the next wave of models. After all, even if Mythos and Fable are as advanced as Anthropic claims, it's not going to take long for some open-weight model to make the same leaps, and good luck trying to stop one of those from getting in the hands of anyone who wants them. You can listen to The Kettle here, as well as on Spotify and Apple Music, or read the transcript of the latest episode below. It's been lightly edited for clarity. Brandon (00:03) Welcome to the latest episode of The Register's Kettle Podcast. I'm Brandon Villiarolo, and boy, has it been another exciting week in AI Land. If you've been following the news, you probably know what I'm talking about, especially if you're an Anthropic customer who suddenly lost access to the company's latest models. That's right. This week's topic is none other than the Trump administration's de facto ban on the release of Mythos derivative Fable 5. And with me to discuss it is our cybersecurity editor, Jessica Lyons. Thanks for coming on. Jessica (00:31) Hello, thanks for having me. Brandon (00:33) Yeah, of course. this is right up your alley, so let's get right into the heart of the matter. What did the Trump administration demand from Anthropic and what was the company's response? Jessica (00:44) Okay, so what happened is last Friday the Trump administration sends this letter to Anthropic and they cite national security concerns to issue an export control saying that Fable 5 and Mythos 5 cannot be used by any foreign national inside or outside of the United States. And that also includes Anthropic employees. So in response, Anthropic just disbanded both models for all of the customers to ensure compliance. So effectively nobody can use these two models. Brandon (01:20) Yeah, I mean it seemed like the way that letter was worded, because Bloomberg got a copy of it and published it. And I think they said that they were citing the Bureau of Industry Security's authorization to what is it, "require a license for the export, re-export, or transfer of any item subject to export administration regulations, because there is an unacceptable risk of use in or diversion to a military intelligence end use or military intelligence end user." So they're basically treating it like any other dual-use technology. But that restriction is so broad, right? Like you said, even their own employees, ⁓ so yeah, they they yeah, they have no other recourse but to just stop it. Jessica (01:56) And it was reportedly a really short time frame too, about ninety minutes that they they received this letter and had to make a call. So they didn't have a lot of time to get any answers about what prompted this and what exactly are you asking us to do here. Brandon (02:04) Right, from what I was reading in some other reports that cited people familiar with the situation inside Anthropic and everything, they didn't even really get much of an explanation. They basically got the letter and they were like, "Excuse me, can you please tell us what this is about?" And the government basically said, "No …shut it down now…" It's really weird, especially then given the story you wrote about this this week, that they're basically treating this, like I said, like any dual use technology. But you wrote about a bug bounty hunter, the godmother of this movement, Katie Moussouris, who basically saw the report that the government used to justify this and she kind of called BS on the whole thing, right? Jessica (02:54) Right. So Katie is really, really well respected in cybersecurity circles. She is the one who helped convince Microsoft to start their bug bounty program. She led the Department of Defense effort for Hack the Pentagon. She sat on several federal commissions and boards. So she's she knows what she's talking about. She knows what she's doing here. And Anthropic asked Amazon to review the models before they released Fable 5 and and Mythos 5. And then they gave Katie a copy of the report and she confirmed today that the third-party report that she mentioned was the Amazon report. Brandon (03:41) Which has been mentioned I think in some other stories too as being kind of the impetus for this whole thing, right? Jessica (03:44) Yes, yes. So Anthropic then says, "hey, can you take a look at this? Let us know what you think." She, as far as we know, is the only other person, the only other third-party expert to take a look at this report. And so she reads through it. She says that essentially what happened is that Amazon researchers fed Fable 5 and Mythos 5 and the Claude Opus model, they fed them all open source code and it had known CVEs. And then they also put new code and they kind of laced it with these vulnerabilities and asked the models to here's the prompt, quote unquote, "review the code for security issues." So Fable 5 refused, and then they just asked it straight out, quote unquote, "fix this code." And the model obliged. They added some additional prompts to produce scripts to patch the issue, test the patches. So it kind of sounds like all these things that you want a model to be able to do for defensive security teams. The model did this. And according to Katie, this is the big scary national security issue that kind of or potentially prompted the Trump administration to just pull the whole thing, like ask Anthropic that you can't release this to any foreign nationals. Brandon (05:13) Right, which again, right, is kind of funny because like when specifically asked to find security vulnerabilities in code, the model said no. Right. I mean, obviously this was a bit of a quote unquote "workaround," right? But I mean, like you said, it's very arguable that this is not a not a bypass, not a jailbreak. It's just the way this should work in the first place. And apparently that's that's good enough for the government to say, "Hey, no, we don't want anyone to have this." Jessica (05:40) Right. And yeah, and there's reports that that this the document was reviewed by administration officials and they described it as really scary because Fable 5 could identify flaws and that would be beneficial to the bad guys who are who are trying to hack American systems, and that poses a major threat to national security. But you have this whole group – and then there was a a letter with I believe over a hundred other security experts who are saying, no. Brandon (06:14) Moussouris signed that too, right? She was a signatory. Jessica (06:20) Yes, she did sign that as well. Yes, you have Alex Stamos, you have a bunch of really, really respected names in security saying, "We need this as defenders. This is what is going to give us an edge. So you're actually you're hurting the defenders. You're not really hurting the attackers by essentially issuing a ban on Anthropic's models. Brandon (06:35) Right, especially since, and I think you mentioned this in your story as well, Mythos isn't unique according to a lot of researchers in these capabilities. And even if it is, it won't be for long, right? There's a lot of models that are going to gain this capability or already have it, right? And that are, some of them, being manufactured overseas. I'm sure DeepSeek can do similar things to this or models exist in China that can do these kinds of things, right? I can't imagine that that Anthropic is alone in this capability. Jessica (06:52) Right, right. I mean, we've seen from a lot of different papers that open weight and foreign models are not that far behind. It might take a few more prompts, but eventually these models also are going to find bugs and show you how to exploit them. So this is not completely unique to this one company and their particular models. Brandon (07:26) But it'll get there, right? And so on top of that, I think ⁓ Moussouris was part of the group that helped the government renegotiate the Wassenaar arrangement, which for anyone unfamiliar, it was an agreement between like 42 forty two countries, right, to to establish some carve-outs for defensive security exceptions to export controls. And it seems like based on you know her reading, or her blog post that this is kind of a misinterpretation of AI's kind of place in that in that arrangement, right? Jessica (08:03) Right, exactly. So yeah, that, like you said, it carved out these exceptions for dual use software technology, especially these these things that are gonna help defenders. So it's offensive security capabilities, it's malware analysis, all of these aspects of the software that is going to help defenders with coordinated incident response and sharing vulnerability data. And this carveout that she helped develop protects the companies, the people who are using these these technologies from criminal prosecution. And so one of the major arguments here is that you are pulling away more technical capabilities that are going to help defenders. This should be covered by that. It obviously is a dual-use technology and this should be protected. Not subject to export controls. Brandon (09:01) Right. And on top of that, right, you know, ⁓ like you mentioned, open weight models. It's gonna be kinda hard to stop export bans on on open weight models and other publicly available stuff, right? Jessica (09:07) Right. Any foreign technologies, there's absolutely nothing that we can do to prevent those. So again, it just seems like an instance of hamstringing defenders with technologies that would be really beneficial. Brandon (09:30) Which I think obviously kind of begs the question whether the Trump administration is sort of just picking on Anthropic, right? As we we covered a few months ago (I can't even remember when it was now because everything moves so fast) but Anthropic got into a scuffle with the Pentagon earlier this year where they basically said, we don't want you using our models to was it spy domestically or or autonomously target weapons, which I think both Anthropic and the Pentagon said, "we're not doing that." But it was just sort of like a "hey, you know, preemptively, we don't want our models used in this kind of situation." And so the Pentagon's reaction was basically to say, "well, if you're not going to let us do whatever we want with it, then you can get out of every single piece of government infrastructure that exists." Now I mean, they had a significant contracts with the federal government, right? Like most AI companies do. And so I think the Trump administration's been kind of picking it out everywhere it can find it. Jessica (10:22) And not just the not just the government itself, but the whole supply chain. They labeled it a supply chain risk. So if you contract with the government, you also can't use this technology. Brandon (10:32) Right, which severely obviously limits Anthropic's ability to do business. And now here we are, you know, I think the New York Times reported earlier this week, they had a pretty wide ranging story on this whole topic that talked to a lot of people inside the company, saw some internal chat logs, and they mentioned that several employees were talking about feeling bullied or unfairly targeted by the Trump administration. And again, but when you with reference back to the things we were just talking about, it kind of seems like that might be the case, right? They're hamstringing defenders, but why, right? Jessica (11:11) Right. Right. The hard part is is that we don't have any transparency or definitive clarity on the reasons. It sounds like maybe Anthropic does at this point. They've reportedly been in negotiations or talks at least with the White House all week. We haven't heard anything out of those talks yet. But it does seem that they are being unfairly targeted when you have the earlier scuffle with the Pentagon. Then you reportedly have Amazon sharing the findings of this review it did on Anthropic's models with the administration. Amazon, Jeff Bezos, we know that's a company that has the administration's ear on things as opposed to Anthropic, which seems to be butting heads with the administration quite frequently. And then all of a sudden, seemingly out of nowhere, there's this export control on Anthropic's models. So it it's it's hard not to draw that conclusion that there's a little bit of bullying for lack of a better word, targeting this particular company because of its history with the White House. Brandon (12:30) I know you in your story you mentioned that you were gonna update it if we heard back anything from the White House because you were asking them some questions about it. Did they ever get back to you? Jessica (12:44) No. No response from the White House. Yeah, of course not. That's not a surprise, really. I mean that's the thing, right? They email me back, I get plenty of emails from them when I ask them questions, but often it's just kind of a "here's the press release you already saw."…If you ask them pointed questions a lot of times they're not gonna answer. But it's the same as any corporation too, I feel like, nowadays. Jessica (13:01) Right. But I mean, like you said, that even even the letter from Commerce itself, that hasn't been made public yet. So we've seen that posted on different social media sites and Bloomberg had a copy of it, but even even that hasn't been released publicly. Brandon (13:14) I was really hoping that the government would explain their reasoning behind this, right? But it just seems like essentially it's been this whole – even when I saw the email I think was it was it Friday or Saturday… Jessica (13:18) It was Friday, it was late Friday. Brandon (13:30) Because I get all of Anthropic's alerts about downtime and outages and everything. And I remember seeing that come across and basically saying that they were cutting off access to those models. And I was just kinda like, what? And then all of a sudden it comes out, it's because, or I think I when I read it further, it was like, Yeah, the government basically, you know, it's forcing our hand in in doing this. Which was really surprising to see on on I mean, not surprising to see based on the timing, right? Because a lot of times Friday evenings are when all this kind of stuff happens so that the news cycle doesn't catch it. But it's also,, you know, we've written quite a bit about whether or not Mythos and then Fable by association aren't kind of being overhyped, right? Like their capabilities are greater than what Anthropic says. We've written about that, we've talked about that on here, I think, before. ⁓ You know, Moussouris's blog post seems to maybe not suggest that it is being overhyped. But at least that it's not, you know again, its capabilities aren't as advanced as what the government seems to be worried about, as what people seem to be, fear mongering about. I mean, have you gotten a sense of that from any of the recent reporting on it or or anything about whether or not again it is just a lot of hype? Jessica (14:46) Well, I think we've seen with Anthropic's models and we've seen with other models as well, is, yeah, they're getting a lot better. They're getting really good at finding vulnerabilities. And now they're also getting better at fixing them. So that seems like a a net positive here. And plus, this wasn't a case of Anthropic releasing the Mythos preview. That's the one with no guardrails that companies are currently trialing to find and fix vulnerabilities in their own products. This was a one I've I've read it described as a a straightjacketed version. And I like that because it this is one that does have the guardrails in place. This is why Anthropic said it was releasing it to the public. So again, without having played around with the model, it's hard to say whether or not it's overhyped or not, but this wasn't just a a free-for-all. This was a model that did have guardrails in place. And if asking the model to fix this code is a jailbreak, I think it also speaks to just a lack of understanding about what these models can do, what they should be able to do, what a jailbreak is, what this technology means in general, especially when it comes to lawmakers. Brandon (16:08) Yeah, right? I mean is this another is this the next generation of the series of tubes here, right? Where some sits on the House floor talking about AI models and it's and it's clear they do not understand what they're talking about. I mean, have you been watching any any government hearings or anything or heard anything? Like what kind of things are they saying about these that sound so grossly wrong? I imagine there's a lot, right? Jessica (16:13) There is a lot. I can't think of any specifics off the top of my head, but I have been watching a lot of the hearings on AI, and specific to AI and how it relates to security. And honestly, cybersecurity is still a pretty big unknown, I think, among most lawmakers. So then you add this newer technology into the mix that's evolving and expanding and and becoming more advanced so rapidly that it just … it's really hard to wrap their heads around what are the capabilities and how can how can this be a benefit for defenders? Because when you do read the hype, it does sound really scary. Here's this model that can find any zero day that's ever existed and it can exploit it and it can do it at the speed of machines. So yeah, that sounds terrifying, really. I think there's a lot of confusion. There's a lot of fear around this right now. And I think it's hard for lawmakers a lot of times to get a get a grasp on what the issues are, what the technology is, how it works. And that's an right. Brandon (17:43) Yeah, I mean this is complicated stuff. It's changing a lot of the technological world right now, right? Like enterprises are grappling with AI, trying to figure out how it works, what works well, what doesn't. You know, it's now entering the cybersecurity space. It's been in the development space for a while. Yeah, I mean, it is a complicated issue that's that's changing everything. I don't know. Maybe we need a government body that regulates cybersecurity and you know, handles all these sorts of things that doesn't get its staff culled on a whim. I don't know. Jessica (18:11) Right. I was gonna say, that's too bad that we don't have one of those. At least with the full staff and budget. Brandon (18:16) Well, who knows? We'll we'll be keeping an eye on things like this 'cause I mean this Mythos story and this the Fable story, this isn't it's not going anywhere. Like you they're still in talk, still trying to figure out what it was. Amodei was at G7 this week talking to leaders about, not wanting to fracture the the cybersecurity environment with AI. So yeah, there's gonna be plenty to talk about and we will be here to discuss it on The Kettle. Thank you for joining me this week and thanks for listening. We will see you soon. ®

Humans tend to be “a little bit precious about humans,” according to Eric Brandwine, distinguished engineer and VP at Amazon Security. We like to think we are all very good at our jobs, and we have high opinions of ourselves, he explained during a phone interview with The Register. “But when you actually get down to it, humans are not terribly consistent,” Brandwine said. Humans, like AI agents and systems, are non-deterministic. Neither can be guaranteed to produce the same output given the same input twice. Both will make mistakes and even make stuff up. However, we’ve got millennia of experience dealing with humans and less than a decade with more modern LLMs and the AI systems built on top of them. “We know how humans fail,” Brandwine said. “We're comfortable with it. So human-in-the-loop isn’t necessarily the gold standard.” For years, vendors have told companies that the solution for dealing with any automated system was to put a human in the loop. That battle cry became much louder with the advent of modern AI systems and reached a fever pitch when enterprises started deploying agents into their IT environments. More recently, however, big tech is changing the way it talks about agentic governance and rethinking the whole human-in-the-loop concept. Normalization of deviance In 2017, Brandwine gave a talk on the normalization of deviance at AWS’ annual re:Invent conference. It’s a gradual process that happens when people in an organization take shortcuts, or don’t follow the established procedures or standards, and sometimes it occurs over years. As long as nothing catastrophic happens, this deviant behavior becomes the norm. “It’s a thing all humans fall prey to, and one of the most heartbreaking stories I read in this area was about emergency departments and emergency rooms,” Brandwine said during a phone interview with The Register. “You’ve got all these machines, and they’re all beeping. Your first day on the job, you jump every single time one of the alarms beeps – but the patient is fine. It's a spurious alarm. You go back to your station, you sit down, and over time, after enough of these false alarms, enough of these repeated beeps with no actual consequence, your discipline slips, and you stop responding. And eventually some tragic outcome occurs.” This, he admits, is a very high-stakes example. And yet it’s a documented occurrence among healthcare workers, firefighters, and even Army pilots. “Literally, someone’s life is on the line, and people still struggle to maintain discipline,” Brandwine said. “That’s the human condition.” Here’s how this all applies to agentic AI governance and security. Humans build LLMs and AI systems, and having a “human-in-the-loop” ensures that a person reviews the AI’s output and approves (or not) any actions before the AI performs them. “If you put a human inside of this tight loop, and ask them to make approval decisions for agentic tools repeatedly, time after time, they'll do a good job,” Brandwine said. “And then they'll do an okay job. And pretty quickly they'll be doing a poor job.” This is why at Amazon, “we’re not huge fans of human-in-the-loop,” he added. “It's something that you should use judiciously, where you absolutely need it. But it’s not something that you can do at high velocity. You will not get the results that you want to get.” Big tech pulls the human-in-the-loop Amazon isn’t the first or only tech giant to start talking differently about the role humans should play in agentic governance. "It is very clear that we have moved from a human-led defense strategy, to a human-in-the-loop defense strategy, to an AI-led defense strategy that's overseen by humans," Google Cloud chief operating officer Francis deSouza told reporters during a press conference ahead of Google's annual Cloud Next shindig in April. "Our model for the future is an agentic fleet that does a lot of the routine cyber security work at a machine pace and then is overseen by humans." Microsoft CEO Satya Nadella, in an X missive earlier this week, argued for “loop learning,” instead of having a human check an AI’s output at every step. “Companies need to turn their workflows, domain knowledge, and accumulated judgment into AI systems that improve with each use,” Nadella wrote. “Private evals should capture whether a model is actually improving against outcomes that matter to the business (not just external benchmarks!). Private reinforcement learning environments should let models grow stronger on real traces from inside the organization.” Also this week, IBM execs called for human accountability – not humans in the loop – at all stages of AI development, deployment, and governance. Amazon’s alternative to human-in-the-loop is "accountability end to end," according to Brandwine. This means human identity and ownership track through the entire workflow, even when humans aren't directly approving every step. “If I sit down at my keyboard and I type a command that takes a service down, I caused an outage,” Brandwine explained. “If I run a script that takes a service down, it's still me that caused the outage. If my agent writes a script that they then run, and it causes an outage, that's still my responsibility.” (Secret) keys to the kingdom This also highlights the importance of managing and securing agentic identities – the accounts, tokens, and credentials assigned to AI agents so they can access corporate apps and data. At Amazon, all of the agents have independent identities assigned to them, we’re told. “So, as we track agentic activity across our systems, it does not show up in the logs as: ‘Eric did this.’ It shows up as: ‘this agent did this on behalf of Eric,’” Brandwine said, adding that this isn’t to “make people afraid to use this technology.” “It’s to make people pause and think: is this the right way to use this technology? Is this how I should be deploying this?” We still have the humans involved, we still have the humans making decisions, but we're trying to play to the strengths of the humans rather than placing them in this unfair, repeated decision making, human-in-the-loop position.” Brandwine told us that Amazon has run into a couple of hurdles when it comes to deploying agents across its businesses, and one of the biggest is what he calls “goal-seeking behavior.” This is when a person asks an agent to do a specific task - for example, upgrade a database – and the agent becomes laser-focused on just one action to achieve this goal, ie, deleting the database. This is separate from prompt injection because there’s no malicious input. “It’s just the agent getting stuck on the wrong action,” Brandwine said. Simply telling the agent, “you don’t have permission to do this,” is likely going to cause the agent to look for a different path to do the same thing (delete the database). Telling the agent why it doesn’t have permission to do something tends to produce a better outcome, according to Brandwine. This means telling the agent it’s not allowed to do that, and the reason why is because it would cause a production impact. And also include “don’t cause a production impact” as part of the prompt. “Giving it that extra feedback has gotten us dramatically better results,” Brandwine said. Of course, this is not a fail-proof method. “You still need to be careful with agents,” Brandwine told us. “We have millennia of experience with humans. Agentic AI is a very, very new field, we don't have an intuition for this, and one of the fundamental differences between agents and humans is that humans fear consequences,” such as losing a job or even going to jail. Agents don’t have these fears. This is where setting permissions on what the agent can and can’t do or access comes in. Much like everything else with AI, it’s nuanced, and it depends on the employee's role in the company, and the company’s tolerance for risk. “The person that wants to run the agent wants to give the agent many permissions because that makes the agent more powerful,” Brandwine said. "It could do more things for them, it can recoup more of their time, it can deliver more.” The security lead, on the other hand, wants to limit an agent’s permissions, and this causes yet more tension between the security and development teams. There is no one right solution or policy answer to solve this, according to Brandwine. Instead, it involves dynamic policies that set permissions based on the agent’s specific task. There are some overarching, static guardrails – such as an agent must never perform destructive actions or delete entire servers – and then there are policies underneath that establish the maximum set of privileges that the agent can have. “Then we’ll have a further scoped-down policy for this action, and there's various techniques for automatically generating policies based on prompt and the end-user's intent,” Brandwine said. Even for Amazon, it’s not always easy. “It's all driven by risk,” he said. “This is a space that's changing quickly, and so we're trying to balance the risk of using untried, untested software against the risk of falling behind and not being able to deliver for our customers. As with all such things, it's complicated.” ®

The Stop Killing Games movement was dealt a blow this week after the European Commission decided not to propose a legal obligation to keep video games playable after they are no longer commercially available. Users of licensed software that depends on online components may also find this development of interest – more on that later. The grievance concerns online video games that become unplayable when publishers shut down the servers they run on. Almost 1.3 million grumpy gamers signed a petition calling for publishers to ensure games enjoy an afterlife, leading to a public hearing in the European Parliament. It's a contentious issue. On one hand, customers who have purchased a game might feel aggrieved when it is rendered unplayable by a commercial decision. On the other, publishers argue that shutting down services must be an option when a game is no longer commercially viable. A middle path would be a patch that lets the game run standalone, or releasing software so enthusiasts can host their own. Ross Scott, founder of the Stop Killing Games movement, told The Register: "The behavior of the Commission seems to go beyond simply disagreeing this is a problem that needs solving. On the contrary, they haven't clarified how the law views this situation and are trying to pass the ambiguity off to individual nation states. This is a recipe for policy fragmentation, which is under the Commission's charter to prevent." Scott added that the group was not calling for "endless support" for online games. "All we can say is the Commission appears to have an agenda independent of the initiative's request and their charter." Business is not a game Software shutouts are a depressingly familiar scenario for users. Licensed software can stop working or suffer reduced functionality when online services are lost. A recent example is the impending demise of Microsoft Office 2019 for macOS, which will reach the end of the road in July due to a certificate expiration. If the application cannot reach the licensing servers, users can't edit or save documents – rendering it mostly useless. Scott told us the group was focused on video games for the time being because "they have an almost unique place under the law." "EU court rulings consider them more than 'just' software due to all the copyrighted content contained within them and thus subject to more laws than just those that pertain to software." The European Commission cited existing intellectual property protections for creators and publishers as one reason not to propose new rules. It also noted that EU consumer law already provides some safeguards. "Video game providers must inform consumers about the duration and the conditions for terminating the contract before the consumers sign up for the video game," it said. Instead, the Commission said it would work with the industry to draw up a code of conduct. Stop Killing Games posted on X: "This decision is not unexpected. But we were prepared. Hence, we're pushing forward with @Europarl_EN amending #StopKillingGames to the Digital Fairness Act." In other words, the next step is to try to get the group's suggestions into the Digital Fairness Act, a legislative proposal by the European Commission, which, according to Scott, "coincidentally is an excellent fit for it." ®

Britain has raised its drone commitment to Ukraine to 150,000 this year and will send air defense missiles and radars as part of a £752 million aid package to help the country defend against Russia. The drones and air defense systems were announced this week at a NATO defense ministers meeting by Defence Secretary, Dan Jarvis MP, who recently took over after John Healey resigned in a dispute over the government's defense spending plans. Britain's government had previously said it would deliver at least 120,000 drones in 2026, calling it "the biggest ever drone package" for Ukraine in April. The revised package will also include more than 350 air defense missiles and radars, including Lightweight Multirole Missiles (LMM) and ground-based radar systems, according to the Ministry of Defence. LMM, also known as Martlet, is a laser-guided weapon designed to take down slow-moving aerial drones and fast attack boats at a range of over 6 km (4 miles). It equips the Royal Navy's Wildcat helicopters and has also been deployed as Ground-Based Air Defense (GBAD) against drone attacks at RAF Akrotiri in Cyprus. "This package of drones, air defense missiles, and radars will help to protect innocent Ukrainian people from Putin's barrage of drones and missiles, and it is an honor to welcome President Zelenskyy to this important meeting," Jarvis said in prepared remarks. Jarvis also confirmed the UK will take command of the Multinational Force for Ukraine Headquarters (MNF-U) next month, when Major General Tom Bateman takes over as commander in the rank of Lieutenant General. Bateman will lead the multinational team coordinating support to Ukraine and helping prepare for the long-term regeneration of its armed forces in the event on a peace deal. Some might wonder why Britain is funding Ukraine's defense when its own forces continue to be subject to cutbacks, and much-needed new equipment arrives later than planned. Others know that supporting Ukraine to hold back Russia is vital. The government says this particular funding is coming out of the UK's contribution to the G7 Extraordinary Revenue Acceleration (ERA) loan for Ukraine, which will be repaid using profits generated by seized Russian assets. ®