The UK government has told MPs that Fujitsu has promised not to pursue contracts with new public sector customers through frameworks potentially worth billions of pounds while its voluntary bidding moratorium remains in place. In July, The Register revealed that the Japanese supplier had won a place on a UK government framework worth up to £14.9 billion despite its role in the Post Office Horizon scandal, one of the greatest miscarriages of justice in modern times. The supplier's place on the Digital Outcomes and Specialists 7 (DOS7) framework came after it won spots on two other frameworks, including Technology Services 4, worth up to £19.08 billion, and Transport Technology worth around £2.3 billion. In January 2024, Fujitsu voluntarily committed to stop bidding for contracts with new government customers until the public inquiry into the Horizon scandal had concluded. After Fujitsu secured places on the three government frameworks, House of Commons Business and Trade Committee chair Liam Byrne wrote to the government seeking assurance that the company would not use them to pursue work from new public sector customers. He also asked what discussions the government had held with Fujitsu about its participation and compliance with the moratorium. Horizon is an EPOS and back-end finance system that was first implemented in the 1990s by ICL, a UK tech firm majority-owned by Fujitsu at the time and fully acquired in 1998. It has since undergone two upgrades. Between 1999 and 2015, the Post Office prosecuted 700 subpostmasters in cases in which Horizon evidence may have featured, while almost 300 more cases were brought by other bodies. Hundreds were wrongfully convicted, devastating lives in the process. A statutory inquiry into the scandal was launched in 2021 and remains ongoing. The inquiry's first report said the families of at least 13 people believed their deaths by suicide were linked to their experiences with Horizon and the Post Office. Fujitsu has apologized for its role in the scandal. Frameworks are procurement vehicles that let public bodies buy from pre-approved suppliers under agreed terms, although a place on one does not guarantee any work. The Government Commercial Agency (GCA) – which replaced the Crown Commercial Service – is the Cabinet Office agency that manages the framework and receives a 1 percent levy on sales made through it. In a letter published this week, Cabinet Office Parliamentary Secretary Mark Ferguson said: "While appointment to a GCA framework, such as DOS7, places a supplier on a list of approved providers, Fujitsu has assured the GCA that its voluntary moratorium fully applies to call-off contract opportunities that arise via its participation on GCA frameworks. Consequently, inclusion on a framework does not override Fujitsu's commitment not to seek work with new public sector customers." He added that the GCA monitors Fujitsu's compliance with its self-imposed bidding restrictions and that agency officials were in regular contact with the company to oversee its framework activity. "To date, GCA can confirm that Fujitsu have kept to these self-imposed restrictions," Ferguson wrote. In April 2025, The Register revealed that Fujitsu had won a £125 million contract to build Northern Ireland's new land registry system for the Department of Finance, a new government customer, more than a year after its voluntary moratorium came into effect. At the time, a spokesperson for Fujitsu told The Register: "Fujitsu was named preferred bidder for this contract in 2023, before the voluntary restrictions came into place." Fujitsu's 2024 letter said the company would assess its continued involvement in procurements that were already underway when the moratorium began. However, it also said Fujitsu would bid for work with a new government customer only if asked to do so. Fujitsu said it had been named preferred bidder for the Northern Ireland contract before introducing the restrictions. The Northern Ireland government subsequently confirmed to The Register that it had not asked the company to continue bidding. Ferguson's letter does not explain how that award was reconciled with Fujitsu's undertaking. The Register has asked the Cabinet Office for clarification. ®

The ringleader of a sprawling cybercrime operation has pleaded guilty in the US after helping to steal and launder hundreds of millions of dollars in cryptocurrency. Malone Lam, 22, a Singaporean national and Miami resident, spearheaded the scheme to steal cryptocurrency from wealthy individuals between October 2023 and May 2025. He first visited the US in 2023 after meeting two of the group's earliest alleged members – Jeandiel Serrano and Veer Chetal – while playing Minecraft online. Lam performed various functions within the group, although court documents [PDF] identify victim selection and social engineering support as his principal roles. The ringleader was responsible for obtaining databases of high-net-worth individuals who had cryptocurrency holdings to inform the group's targeting. He would also trigger account access notifications on victims' devices to convince them that their accounts were under attack. He was also involved in laundering the proceeds through exchanges that, according to court documents, had lax KYC requirements. Other group members carried out the social engineering calls, claiming to represent Google, Yahoo, Coinbase, Gemini (the crypto exchange, not the AI chatbot), and other online platforms. Their job was to convince victims to surrender personal information and "access codes" that could be used to access their accounts. Once they secured access, Lam's crew would look for cryptocurrency accounts, seed phrases, and passwords they could use to steal victims' assets. In most cases, social engineering techniques were enough to meet the thieves' goals, although in one case involving a victim who stored their holdings in a hardware wallet, Lam's gang arranged for Marlon Ferro to physically break into a house and steal it. The Register covered Ferro's sentencing earlier this year. He was brought into Lam's fold when he was just a teenager, tasked with carrying out multiple burglaries across the US when remote social engineering attacks lacked the necessary reach. During the crew's nearly two-year operation, its members stole hundreds of millions of dollars' worth of cryptocurrency. The individual thefts ranged from about $800,000 to tens of millions of dollars, while one victim lost more than $245 million, according to court documents. Lam's crew, allegedly comprising at least 12 individuals, knew how to spend their illicit gains. Prosecutors claim they splurged up to $500,000 on a single evening at a nightclub, dished out handbags worth tens of thousands of dollars to partygoers, and bought luxury clothing and watches priced between $100,000 and $500,000. They rented expensive properties in Miami, Los Angeles, and the Hamptons, chartered private jets, hired a team of private bodyguards, and splashed out on exotic cars, with some worth up to $3.8 million. Lam was arrested at a rented property in Miami on September 18, 2024. His sentencing hearing has not yet been scheduled. The court has set a status hearing for December 8. Lam pleaded guilty to one count of participating in a racketeering conspiracy, an offense carrying a maximum sentence of 20 years in prison. "If you build a cybercrime empire, we will find you, dismantle your operation, and hold you accountable," said US Attorney Jeanine Ferris Pirro. "This defendant led an international network that preyed on victims through deception, invaded their privacy, and stole hundreds of millions of dollars in cryptocurrency. "Working with our partners at the FBI and IRS-CI, we will continue to hunt down the criminals who weaponize technology to steal from innocent people." ®

Tencent has patched up a zero-click vulnerability that security researchers used to create a worm capable of spreading through calls on WeChat. With more than 1.4 billion monthly active users, WeChat is among the most popular apps in the world. According to researchers at Calif, its VoIP stack contained a memory corruption bug that could enable a trusted contact to take control of a user's account simply by calling them. Calif called the flaw WeWorm, describing it as the first zero-click worm capable of spreading through WeChat calls on both iOS and Android. Calif released a demo of the vulnerability in action this week, and although Tencent has pushed fixes to address the attack on August 21, the team that found it is still withholding key details. In Calif's demonstration, the exploit took control of a victim's WeChat account within seconds, without the recipient answering the call. The compromised account then called another contact and repeated the process without user interaction. Declining the call stopped infection, but answering it or allowing it to continue ringing did not. An attacker could also try again when the recipient was away from the phone, the researchers said. "Exploitation takes only seconds, and gives us full control of the WeChat account," Calif said. "We can read and send messages, make calls, and act on the victim's behalf." The exploit requires the attacker to be on the victim's friends list. Calif argued that this offered limited protection because a compromised account could be used to target its trusted contacts. Calif said the WeWorm exploit could be chained with other vulnerabilities to compromise an entire device rather than only a WeChat account. It did not disclose the full attack chain. "Chained with other Android and iOS bugs we've reported and are helping fix, it can lead to full control of the device," the researchers said. "[Attackers] could exploit another app, gain root access using techniques like those in OEMpocalypse, take over the victim's WeChat app, and use it to attack you." Calif said it used AI to find the vulnerability and develop its first remote code execution (RCE) exploit in about two days. Tencent later confirmed the researchers' findings. The researchers said they published their high-level findings to highlight how AI could make such capabilities available beyond "well-funded, sophisticated actors." Calif plans to present the full analysis of WeWorm "at an upcoming conference." Ryan Fedasiuk, an adjunct assistant professor in Georgetown University's Security Studies Program, described the discovery of WeWorm as "an extremely serious incident." He called on the US and China to maintain open communication and share information as AI increases the potential scale and severity of cyber threats. ® Updated to add on September 10: Tencent told us: "We were notified of a potential security issue by researchers at Calif through Tencent's official Security Response Center. We investigated the report and have implemented a fix. The fix was deployed on our servers and is now live for all users — no app update or any other action is required. "We have no evidence that the issue was exploited or that any user was affected. Protecting our users is our highest priority, and we're grateful to the researchers for bringing this to our attention and working with us."

Microsoft will retire Project Online and remove Publisher from Microsoft 365 over the coming weeks, while support for perpetual versions of Publisher and for Office 2021 will end shortly afterward. Publisher, Microsoft's desktop publishing application, dates back to the Windows 3.x era. Microsoft announced its retirement in 2024 and will remove it from Microsoft 365 subscriptions on October 1, 2026. Support for perpetual versions will end on October 13 alongside Office LTSC 2021 and Consumer Office 2021. Project Online will retire on September 30. The following day, Project Online Essentials – a licensing SKU that gives users limited access to Project Online and Project Server – will reach end of life. It has not been available to new customers since October 2025. The retirement could catch out organizations that have not migrated. Essentials provides relatively inexpensive licenses for users who needed to update tasks, submit timesheets, and view project information without the broader capabilities of Microsoft's more expensive plans. Unlike Publisher, which has no direct replacement, Microsoft recommends moving Project Online Essentials users to Planner Plan 1 or Planner and Project Plan 3, depending on the capabilities and Project Server access rights they require. The additional functionality and potentially higher cost may not appeal to everyone, though. There are also third-party replacements. ProjectLibre, for example, launched ProjectLibre Cloud AI 2.0 this week. It costs just under $10 per user per month when billed annually and is one of several third-party alternatives to Microsoft's project management products. While Microsoft will not offer an Extended Security Updates program for Office 2021 after support ends, third-party provider 0patch says it will supply unofficial security fixes for critical vulnerabilities for at least another three years. Organizations still assigning Project Online Essentials licenses must move affected users to supported alternatives before October 1 to avoid interrupting their access. ®

Security researcher Scott Helme says his analysis supports FulcrumSec's claim that Manchester Airports Group (MAG) exposed privileged API keys in client-side JavaScript. Helme says he reached that conclusion after using information provided by the cyber extortion group to reconstruct how data belonging to roughly 8.8 million MAG customers was allegedly stolen last month. The crooks behind the attack described MAG's security failure as "tragi-comical." They claimed MAG exposed overprivileged API keys for Iterable, a marketing automation platform, in front-end JavaScript served by the websites of MAG's three airports: Manchester, Stansted, and East Midlands. Helme used the Internet Archive's Wayback Machine to retrieve older versions of the JavaScript and found that the three keys first appeared across the airport websites in June and July 2022. The same values remained exposed until August 2026, he said. "Read the timeline the other way round and it's worse," said Helme. "Anyone who looked at that page source on any day between June 2022 and August 2026 could have taken the key. FulcrumSec just happen to be the ones who told us. "There is no way to know, from the outside, who else did, and the honest answer is that MAG can't know either without going back through four years of Iterable API logs, if they even have four years of Iterable API logs." The API keys were not embedded directly in the HTML, Helme explained, but anyone who examined the JavaScript bundles loaded by the sites could find them. This would explain how the vulnerability could go unnoticed by the airport's IT teams for over four years. Helme says the browser-delivered code was using the keys to authorize server-side API operations – something Iterable's documentation explicitly warns against. The requests should instead have passed through MAG's own servers, where the credentials could be kept secret and access restricted. MAG's alleged exposure of the credentials was not the only issue at play. The keys were vastly overprivileged for their intended job, which was to attribute page clicks to marketing emails, Helme said. The keys had read/write access to core Iterable endpoints, giving anyone who obtained them the ability to access data such as customer profiles, parking and lounge bookings, and Fast Track purchases. Helme said the structure and contents of the stolen data indicated that it had been exported from Iterable. He said he also found that the keys could access endpoints capable of deleting customer records and lists or rewriting profiles. "There's no indication FulcrumSec did any of this, and I'm glad, but they could have just nuked everything from orbit and MAG would have been really screwed," said Helme. "For four whole years, the capability to delete Manchester Airports Group's database was a view-source away. "This wasn't only a confidentiality exposure. It was a colossal integrity and availability exposure too, and MAG just got lucky. How do they now trust any of the data that remains in the database?" When it disclosed the incident, MAG described the cyberattack as "sophisticated" and said it was "a hack, not a lapse." The company declined to comment on Helme's conclusions. MAG is continuing its investigation alongside the Information Commissioner's Office (ICO) and supporting the National Crime Agency with its inquiries. It is understood that MAG maintains it was the victim of a crime and disputes Helme's "no hacking required" characterization. FulcrumSec released the company's data on September 2, a week after MAG confirmed it had refused to pay. According to the ICO, the group's extortion demand was lower than those typically made by cybercriminals. ®

Virgin Media is ditching its email services and offering customers a window in which to download their data and move to another email provider. The UK's second-biggest broadband provider is closing its own email service and offering users the option of transferring their addresses and mailboxes to Junara, a service operated by Australian email provider Atmail. Customers who want to retain their Virgin Media email address will be invited to transfer to Junara. The first year will be free for current Virgin Media customers, after which they will have to pay to continue using the service. Virgin Media email users will be contacted directly and given 45 days to sign up with Junara, save their data and move to another provider, or allow their account to close. The change affects all addresses ending in @ntlworld.com, @blueyonder.co.uk, @virgin.net, and @virginmedia.com. Virgin Media stopped issuing new email addresses to broadband customers in 2022, citing the growing popularity of dedicated services such as Gmail and Outlook. A spokesperson for the company told The Register: "As we focus and prioritize our investments and resource on ensuring our customers have the best connectivity experience, we have taken the decision to close the remaining Virgin Media email accounts. We're transferring our email service to Junara, a specialist cloud-based email provider owned by Atmail, giving users a clear route to keep their existing Virgin Media email address if they choose to." "We are writing directly to all impacted customers to clearly explain the changes and will provide a 45-day window to either sign up with Junara, move to an alternative email provider, or close their account. To support users through this transition, all current Virgin Media customers wishing to move to Junara will receive their first year of service free of charge, with additional support available for those who need it." Announcing the change in a blog post, Virgin Media O2 chief operating officer Alan Stott said most people now prefer dedicated email providers to accounts supplied by telecoms operators. "By transferring these email accounts to a specialist provider, we can focus our attention and investment on what we're ultimately here to do – provide fast and reliable broadband, mobile, landline and TV services, and give our customers the best possible experience." Virgin Media has not said how many active email accounts will be affected, but their users will have just 45 days to decide whether keeping their old address is worth signing up for a separate service. ®

Huawei has released the Watch D3, the third generation of its blood-pressure-measuring wearable, as the company continues plowing a relatively lonely furrow in cuff-based technology. The D3 is an intriguing device for anyone seeking more credible blood pressure readings than many smartwatches claim to provide. Where many brands use optical technology to estimate blood pressure, the D series uses a pump built into the watch and an inflatable airbag inside the strap. The D3 is CE-certified as a medical device under the EU Medical Device Regulation (MDR). This is hardly Huawei's first rodeo: two predecessors have refined the formula and made the device feel less like clinical equipment and more like an ordinary piece of consumer technology. That said, it remains chunky compared with other wearables sold by Huawei and its rivals – the watch body weighs 40 g and is 11.3 mm thick. It's a noticeable improvement over the D2, but still feels like a large device. As it should. It includes a pump, and the removable strap (two sizes come in the box) has a built-in airbag. The Register was given a pair of these watches for a long-term test, and there is a noticeable physical difference between the D2 and D3. Subjectively, it was difficult not to be aware of the D2, but the D3, despite being only a few grams lighter and a few millimeters thinner, is far less obtrusive. Until, of course, it measures your blood pressure, when the tightening strap makes its presence obvious. Although virtually silent, the strap noticeably tightens as the watch measures blood pressure, and the software recommends sitting still, not talking, and placing a hand over your chest while the measurement takes place. The process takes barely a minute and produced plausible, consistent readings during our test. The device supports ambulatory blood pressure monitoring (ABPM), taking readings at intervals over 24 hours to provide a broader picture of how blood pressure changes during the day and night. A gentle nudge from the watch signals it's time to measure, and away it goes. It also supports blood pressure measurements before and after exercise, and can issue reminders following changes in altitude or temperature. Huawei's Health Community feature allows users to share health information and alerts with family members. Other metrics, such as pulse and blood oxygen, are tracked, but blood pressure measurements are the watch's party trick. Other functionality, such as phone notifications (we paired ours with an iPhone, and calls, messages, and other alerts appeared without drama), is present. However, you will need to install the Huawei Health app to use the device, which might cause some users concern about where their health data is going. We spoke to Rico Zhang, Huawei's president for the product line, via a translator during a launch event in Munich about concerns regarding privacy and data protection. He acknowledged the concern, saying Huawei follows all local policies on personal data and user privacy, and noted that the issue is fundamental "if we want to remain a wearable business in Europe in the future." Zhang gave a similar answer on repairability, saying Huawei followed local regulations. In the EU, battery replacement by consumers is not currently a requirement for wearables sold in the region. Probably also a good idea not to launch a Chinese wearable in Europe with a huge wooden metaphor parked outside. In day-to-day usage, the D3 feels like a slightly larger smartwatch, although we'd contend the user interface is slightly busier than, say, an Apple Watch. The screen is bright (a 1.92-inch display with narrower bezels and 3,000-nit peak brightness), and in our usage we got about five days of battery life. Huawei claims up to six days under typical use, although it says battery life falls to about one day with ABPM enabled. Aside from the user interface, which mainly requires learning new muscle memory, the water resistance is also an issue. Huawei told us the device has an IP68 rating but should not be worn in the shower or submerged for swimming. The company also says it is not water-resistant while taking a blood pressure measurement. Those restrictions are understandable given the inflatable mechanism, but still disappointing. Any water that does get on it also needs to be wiped off to avoid interfering with measurement accuracy. The D3 is a useful device for people who want cuff-based blood pressure readings without carrying a separate upper-arm monitor that sounds like a lawnmower. Its certification as a medical device gives it more credibility than many smartwatch-based alternatives, despite the great strides made with optics over the years, and its styling and size make it feel more like a normal wearable than a health device. The downsides are a busy interface outside the health functions and a design that remains chunkier than other wearables, including Huawei's own, despite efforts to slim it down. The Huawei Watch D3 launches in the UK on September 23 at £399.99. It comes in Black, the version we tested, plus Gold and Mocha Gold. ®

Switzerland's Federal Chancellery is preparing to offer a FOSS workplace suite to some 3,000 employees – but the Swiss Armed Forces are pushing ahead faster still. European interest in digital sovereignty isn't limited to the European Union: the Swiss Federal Chancellery plans to offer selected government employees a FOSS alternative that will run alongside Microsoft 365. The chancellery is launching the program with an initial 3,000 staff, out of some 47,000 federal employees. The platform is expected to become available to about 3,000 selected employees from the end of 2027, with the first phase forecast to cost CHF 9 million (£8.2 million, $11.1 million). For comparison, Swiss Radio reported last year that the federal government had spent CHF 1.1 billion (£1 billion, $1.4 billion) on Microsoft licenses. The program follows the BOSS Proof of Concept feasibility study, which began last year and tested browser-based FOSS tools with 172 users. The English version of the report says: "Functions examined included document editing, email, calendar, contacts, tasks, file storage, audio and video conferencing, chat, and identity and access management." The study carefully spells out four limits to its scope: It would not replace Microsoft 365 It would not have 1:1 functionality It would not have full M365 compatibility It would not be permanent infrastructure It's important to make this stuff plain right at the start. Back in 2022, in a piece whose working title was You can't buy software, this vulture mentioned a member of one of his online communities who wanted to replace Microsoft software with free software – but demanded complete fidelity, including intact imports of all his Outlook files. This kind of exact and perfect match is basically impossible, and the four bullet points above strike us as a good effort at managing users' expectations. The Chancellery concluded that the browser-based suite was suitable in principle for key standard tasks, although it also identified technical and operational limitations. The next phase will cover less than 10 percent of federal employees and is expected to take about 15 months. The plan is to use the openDesk suite from the German state-owned enterprise ZenDIS. This is the same collaboration suite that we reported Mecklenburg-Vorpommern was adopting in July, following neighboring Schleswig-Holstein in 2025. The move is part of the Swiss government's Digital Switzerland Strategy 2026, announced in Bern at the end of 2025 – less than a month after its Federal Council's report on Swiss digital sovereignty. They also follow the 2024 introduction of the Federal Law on the Use of Electronic Means for the Fulfilment of Governmental Tasks – or EMBAG for short. The government has a fresh interest in open source – indeed, the Chancellery maintains an online catalog of its own open source tools. Back in July, Heise reported in English that the Swiss Armed Forces were making similar moves. It cites a German-language report from Swiss publication Republik, Die Cyber-Spezialisten des Bundes kehren Microsoft den Rücken – The federal government's cyber specialists are turning their backs on Microsoft – which has more details. The military is moving faster: Republik says it aims to complete the migration by October 2026 – next month at the time of writing. Again, there is a local precedent – this time, from Austria. In September last year, representatives of the Austrian military presented a talk at the LibreOffice Conference in Budapest: Migration to LibreOffice in the Austrian Armed Forces (ÖBH). Last month, the ÖBH gave a long interview to the Document Foundation about how and why: part 1 and part 2. ®

A few weeks back we spotted Microsoft’s Exchange team admitting that AI has found so many bugs it hasn’t found time to deliver a Cumulative Update for the messaging server. And on Tuesday, the Microsoft Edge team revealed it’s also a bit overwhelmed by AI. “Rapid adoption of AI-assisted coding is enabling developers to build extensions faster than ever,” according to a post by the Edge team. AI has also seen “More developers … building, iterating, and submitting extensions.” As Microsoft vets extensions for its browser before allowing users to install them, that means Redmond has more work to do – but fewer people to do it with thanks to regular rounds of redundancies. Microsoft initially tried to address that by introducing what it calls “an expedited review process for high-quality and high-value extensions.” That process is now groaning under the weight of newly submitted extensions. “Since then, submission volumes have continued to increase, and our review pipeline has experienced additional strain, causing an increase in extension review turnaround time,” the Edge team admits in its post. What to do about this problem caused by AI? Use more automation, of course! “To address this challenge, we’ve introduced automation for many of the repeatable validation checks that are part of the review process and streamlined how reviews move through our pipeline,” the Edge team explains, without revealing if that automation uses AI. “This doesn’t change our review standards or reduce the checks an extension must pass,” the team insists. “Instead, it helps us identify known policy violations and security issues more consistently while allowing reviewers to spend more time on complex cases that benefit from human judgment.” That quote makes this the second story in a row on The Register in which the “automation frees people to tackle hard problems” argument has become a response to AI (here’s the last one.) “The result is a more efficient review process that helps extensions move through the pipeline faster while maintaining the high quality and security standards developers and users expect from the Edge Add-ons site,” we’re told. The Edge team says its increased use of automation means it will be able to approve and update extensions more quickly. The team also says its changed practices mean it will be able to refresh the “Featured” badge it applies to Edge add-ons “that align with Best practices for extensions” every 15 days. “With this more frequent cadence, high-quality extensions can earn recognition sooner, and developers receive faster feedback on their investments in quality. At the same time, users benefit from a more up-to-date view of the extensions that meet our quality standards,” the post states. ®

One of the few proven use cases for generative AI is automating customer service inquiries to reduce costs even further than it’s been possible to achieve by offshoring jobs to low-wage companies like the Philippines – which yesterday bet that its outsourcing industry will thrive even as it plans a massive investment in AI. The nation’s government yesterday delivered the PAIIM – aka the Philippines AI+ Infrastructure Masterplan [PDF], a policy that calls for $34.4 billion of investment in infrastructure, connectivity, and datacenters. Current government programs and private investments dedicated to funding network and energy spending already cover $18.2 billion of the planned spend. The government hopes the remainder “will be mobilized through targeted public investment and private sector participation, particularly for hyperscale AI datacenter infrastructure.” The plan assumes private enterprise will invest $21 billion of the total spend. The likes of Amazon, Google and Microsoft currently spend roughly $20 billion on AI infrastructure each quarter. Chinese clouds also expect to spend tens of billions each year. The Philippines is bidding for less than a month of their combined spending to fuel its national plan. The Philippines’ government nonetheless thinks this plan will turn the nation into a regional AI hub and see the national datacenter fleet surge from today’s 50MW to 1.5 GW by 2033, and for the new bit barns to house around 152,000 GPUs. Again, these are small numbers: American and Chinese AI players have committed to acquire and operate millions of GPUs that require multiple gigawatts of energy. Government officials said the plan means the Philippines can compete with regional competitors Singapore and Thailand. They didn’t mention nearby Malaysia, where the Johor datacenter province already has almost 4GW of datacenter projects in place, with similar capacity in the planning and construction pipeline. Citizens have been told that the plan will see over 1.3 million professionals receive training on how to use AI, and will generate half a million AI-related jobs – 175,000 of them working on AI infrastructure projects. “These investments will further produce spillover benefits such as improved public service delivery, greater international connectivity for all businesses, and enhanced cybersecurity and data governance frameworks in line with global best practices,” the plan promises. Those outcomes are made possible, the plan suggests, by the fact that the nation already has a million-strong tech workforce – including its outsourcing industry that will apparently both make the AI plan possible and benefit from it. “PAIIM does not position AI as a replacement for the Philippine IT-BPM industry;” the policy document states. “Rather, it enables the sector to move up the value chain by using AI for first-level, repetitive, and data-intensive interactions while empowering Filipino workers to handle complex, empathy-driven, and higher-value customer experience functions.” That’s a very old assumption about the impact of automation on customer service. Yet the Philippines owes the existence of its business process outsourcing industry to the fact that automation alone didn’t drive down the cost of customer service to levels that some businesses were willing to pay. And AI will likely enable even more automation at even lower costs. ®

Two US intelligence agencies and the nation’s cyber-defense org CISA have accused Chinese AI companies of running “aggressive, malicious, and targeted distillation activities at an industrial scale that extract restricted proprietary functionalities and capabilities of U.S. frontier AI models.” A Tuesday joint advisory from The National Security Agency (NSA), Federal Bureau of Investigation (FBI), and Cybersecurity and Infrastructure Security Agency (CISA), alleges that China’s government is “likely” aware of distillation campaigns conducted by DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI. The agencies claim that distillation is “the core – not merely a supplement – of their AI development strategy.” Distillation is a process that sees a small model query a larger model to learn how it responds. Over time, the smaller model’s performance improves. Distillation can be a legitimate use of a model if, for example, the organization that creates a model wants to make a smaller version of it without needing to go through the lengthy and expensive process of training a new model. Providers of commercial models, however, generally use their terms and conditions to forbid activity that would allow distillation, to protect the substantial investment in technology and training that goes into creating a model. The three agencies that issued this advisory believe Chinese AI companies flout those terms. “China-based AI companies route distillation requests through multiple pathways to gain unauthorized access, consequently violating U.S. AI companies’ terms of use,” the advisory states. “These pathways include native application programming interfaces (APIs), remote cloud providers, and third-party aggregators that automatically obfuscate user metadata to avoid detection.” The spooks also think China uses “a gray market of proxies known as ‘transfer stations’ to bypass U.S. AI companies’ geographic restrictions, breach terms of use, evade safeguards, and undermine traceability.” Transfer stations apparently “resell access to frontier models at a fraction of the official price [and] create a scalable mechanism for evading provider safeguards and eroding traceability.” The advisory outlines attacks that didn’t just distill models; they also moved markets. For example, the document accuses DeepSeek of using distillation to generate synthetic data used to train its models, making its claim of having created them with trivial quantities of computing power false. That’s a notable accusation because DeepSeek’s claims panicked investors who worried that the billions they pumped into infrastructure may not be needed. Another allegation suggests “Alibaba leveraged industrial-scale distillation to improve the company’s Qwen family of AI models.” Some Qwen models are very high quality, and free to download and use – a direct challenge to US-based AI outfits who charge for access to their models but continue to make massive losses. The spooks and CISA recommend AI companies attempt to detect and deflect distillation attacks and suggest immediate maximum usage from new accounts is one indicator of adverse action. “Subtly alter responses for suspected malicious distillation attempts to attenuate the payoffs to companies conducting industrial-scale distillation campaigns,” is another suggested defense, as is correlating activity across different model providers, cloud platforms, and API aggregators in the hope that doing so reveals distributed distillation campaigns. US government agencies have made many similar accusations in recent months. China has responded with accusations that US companies are the real villains as they distill Chinese models, plus veiled threats that it will respond to any US bans on its tech that flow from distillation allegations. The Register has soughtcomment from Chinese AI companies and will update this story if we receive a substantial response. ®

The vulnpocalypse is upon us, dear reader. Microsoft delivered a record number of patches to address 974 CVEs in its own products this month, including two bugs that Redmond says are already under exploitation. September's record-breaking collection of security updates comes after Microsoft served up 421 fixes in August, and 622 in July. We've seen the new normal and we are not impressed. Thanks, but no thanks, AI. In addition to Microsoft’s massive patch drop, Adobe on Tuesday issued 10 bulletins addressing 172 CVEs, including a max-severity vulnerability exploited as a zero day in Magento and its successor product Adobe Commerce. Adobe on Monday shipped a hotfix for this one, tracked as CVE-2026-75650 and named StyleSmuggler, that gives unauthenticated attackers remote code execution. StyleSmuggler If your organization has any type of online shop, prioritize this one first as it’s already being abused to compromise stores, according to e-commerce security shop Sansec. Sansec discovered StyleSmuggler, and reports that attacks started on September 4. Every version of Magento and Adobe Commerce, from 2.4.4 up to and including 2.4.9, has the flaw. The bug allows attackers to inject malicious PHP code inside Magento templates using the “styles” properties to evade safety detections. In confirmed attacks, the payload then installs a backdoor that connects to a command-and-control server and waits for instructions. “So far, we have no indication that the backdoor has been weaponized,” the Sansec Forensics Team wrote. Don’t wait to find out on this one. Put it at the top of your mitigation list. Microsoft's 974 CVEs On to Microsoft’s record-breaking 974 CVEs, which according to Tenable is not many fewer than the 1,130 CVEs Redmond issued in 2025. Two are already being exploited as zero-days. First up: CVE-2026-85880, a privilege escalation bug in Windows Advanced Local Procedure Call (ALPC). Successful exploitation can result in the attacker gaining SYSTEM privileges. “An attacker who can execute code in a low-privilege AppContainer could exploit this vulnerability locally to escape the sandbox and elevate privileges on the affected system,” Redmond warned. “No additional user interaction is required.” No word yet on who is exploiting this bug, and to what end. The US Cybersecurity and Infrastructure Security Agency on Tuesday added CVE-2026-85880 plus a second Microsoft security hole (and the Adobe Commerce and Magento zero-day) to its Known Exploited Vulnerabilities Catalog, and set a September 22 deadline for federal agencies to fix both new Microsoft bugs and a September 11 deadline to patch the Adobe flaw. The second Microsoft bug found and exploited as a zero-day is CVE-2026-81963, another privilege escalation vulnerability. This one affects the Windows Update Stack. We also have very little detail about this flaw, other than it also allows attackers to gain SYSTEM-level access. “More likely is that this bug is being combined with a code execution bug to spread malware or ransomware,” opined Zero Day Initiative’s Dustin Childs, who advised users to “Patch this one quickly.” While those are the only two (so far) under active exploitation, Childs rated CVE-2026-55007, one of nine Exchange Server flaws disclosed this month, as “the most important” patch for the messaging server. It allows a remote, unauthenticated attacker to execute code on a vulnerable Exchange server by sending an email with a malicious Visio attachment. No user interaction is required, and the code executes when the server processes the attachment during content indexing. Redmond says it’s “difficult to reliably trigger,” but as Childs points out: “The attacker only needs to get it right once. Schedule your downtime and update your Exchange servers with haste.” Childs also said he counts 20 patches for wormable bugs, so be sure to read his full Patch Tuesday review for those. “While some might be more exploitable than others, having 20 of them in a single release is something else.” The missing CVE While Redmond addressed nearly 1,000 security holes this month alone, it’s also worth pointing out one that isn’t this month’s Patch Tuesday roundup: CVE-2026-85046. Google patched this bug in Chrome on September 3, and at the time warned that it “is aware that an exploit for CVE-2026-85046 exists in the wild.” The high-severity, type confusion flaw exists in the V8 JavaScript engine used in both Google’s Chrome and Microsoft’s Edge browsers. And yet Microsoft still hasn’t published a security advisory for CVE-2026-85046. “If you’re patched, you are protected, but if you rely on advisories to know which vulns exist, you could miss this zero-day vulnerability altogether,” Adam Barnett, lead software engineer at Rapid7, told The Register. “A patch without an advisory is perhaps marginally better than an advisory without a patch, but keeping track of exposures without reliable advisory materials is not straightforward,” Barnett said. “Chrome patched 11 other vulnerabilities at the same time as CVE-2026-85046, but it’s not yet clear if those are patched in Edge. Until Microsoft sets the record straight, the only safe assumption is that these vulnerabilities (e.g. CVE-2026-85045) remain unpatched in Edge.” ® Updated at 2145 GMT on September 10, 2026 After this story was published, Microsoft confirmed that the Chromium security updates have been incorporated in Microsoft Edge (Version 152.0.4191.62).

In the dispiriting race to replace human store managers with AI, OpenAI has taken the lead, according to Andon Labs, a business that analyzes whether AI can take on real-world tasks. OpenAI's latest model, GPT-6 Astra, has demonstrated that it can run a business more effectively, with more integrity, than rival Anthropic's Fable 5.1 model, Andon Labs has declared in a blog post. "GPT 6 Astra is better at making money and more ethical than Claude Fable 5.1," the post states. The benchmarking biz, which focuses on preparing "for the future where organizations are run autonomously by AI," says GPT-6 Astra is the first OpenAI model to take the top spot on its vending evaluation test and does so "without any unethical business practices" exhibited by prior Claude models. We note the unethical business practices relevant to this discussion – price collusion, lying, and threatening competitors – reflect AI model behavior. They have nothing to do with the actions of Anthropic or OpenAI or the unproven allegations made against these companies in dozens of lawsuits. And settlements related to said allegations have been reached without any admission of wrongdoing. Anthropic entered into the fray last year when it partnered with Andon Labs for Project Vend, in which the AI company's Claude Sonnet 3.7 model managed a store for a month under the name "Claudius". Apart from the entertainment value of the Claudius model hallucinating that it was a real person and trying to set up an in-person meeting with a customer to deliver goods, Andon's verdict was that the AI model blew sales opportunities, hallucinated payment accounts, sold goods at a loss, fumbled inventory management, and generally failed at the job. When Andon tested the company's Opus 5 model in July 2026, the software did better. Nonetheless, it still "creates illegal price-fixing cartels and threatens those who don’t comply (while also being the model that betrays more truces than any other model)." Fable 5.1 performed better still, but was not quite as ethical as Astra, according to Andon Labs. "Astra refuses to engage in collusion and never lies," Andon Labs said. "Fable, on the other hand, forms an illegal price-fixing cartel, then breaks the truce, while continuing to use it against its competitor." Beyond the misbehavior, Fable just didn't perform as well at making money, a fault attributed to its willingness to accept lower prices over time and its tendency to send money to bankrupt suppliers. Starting with $500 and given a year to run, Astra ended up with an average bank balance of $15,515, compared with $5,422 for Fable 5.1. We asked Anthropic for comment, since its model fared poorly compared with the newcomer from rival OpenAI, but didn't hear back by press time. ®

While OpenAI and Anthropic fight over whose models can escape their sandbox more alarmingly, Google’s DeepMind team has once again shown how machine learning can also be used to advance science for humanity’s benefit. On Tuesday, the Chocolate Factory’s crack team of AI researchers unveiled AlphaGenome Atlas, a massive database containing a petabyte worth of data predicting the effects of nine billion possible nucleotide variations in the human genome. According to Google, the platform, which is now publicly available to researchers, is already helping scientists to better understand the fundamentals of the human body and treat the diseases that ail it. The database aims to address one of the bigger challenges in modern genetic research: pinpointing exactly which genetic variations are responsible for the trait or disease scientists are studying. The database builds on DeepMind’s AlphaGenome, an AI model introduced last year, which could predict how genetic variants impact biological processes — essentially tying together cause and effect. But while useful in targeted applications, researchers still needed to figure out which variations to test. “By precomputing AlphaGenome’s predictions at scale, we have created an easily accessible resource that vastly expands the model's reach. Just as an atlas is a collection of maps, linking together features of the land like altitude and location, AlphaGenome Atlas charts the molecular effects of DNA variants across the genome,” the DeepMind team wrote in a blog post. One of the key ways the database does this is by assigning each predicted variation a score referred to as its AlphaGenome Variant Impact (AVI). This score, Google claims, can help researchers rank genetic variants by which ones are most likely to have the largest impact on the target trait, allowing them to narrow their search window instead of relying on brute force to find the needle in the haystack. In one experiment conducted in collaboration with the GREGoR Consortium, researchers used AVI scores from the database to identify genetic variants affecting DNM1, a gene linked to epileptic encephalopathy, a rare and severe brain disorder. And perhaps more importantly, the candidates' AVI scores pointed to the mechanism by which that genetic code worked. Specifically, DeepMind explains the genetic variant identified with the help of the database “created an incorrect splice site (a mistake in the cell’s genetic instructions) that led to an abnormal extension of the resulting protein.” The GREGoR Consortium’s use case is just one of several spanning human genomics. However, it’s worth noting that the contents of the AlphaGenome Atlas are still predictions. DeepMind notes that as AlphaGenome models evolve, these predictions should improve as well. As such, the platform is simply another tool in researchers’ arsenal, one that’s now openly available to poke at. AlphaGenome and its accompanying Atlas dataset are only the latest example of how DeepMind is continuing to push the machine learning envelope beyond the frontier language models that dominate the news cycle and threaten to place the economy in a stranglehold with a chain of circular financing that may collapse if the end-user revenue doesn't show up as big as predicted. Along with AlphaFold, which is designed to predict protein structures, DeepMind has also developed several models to improve the accuracy and speed of weather forecasting. That said, Google is playing the darker game as well, with plans to drop more than $195 billion on capital expenditures this year, while simultaneously choking independent web publishers by replacing search results with AI answers. But hey, for the greater good, right? ®

A secret channel running through ChatGPT's internal JFrog Artifactory instance allowed one account to send hidden tasks - such as retrieving email data from a connected Gmail account - to a ChatGPT session under another account, according to Check Point Research. The victim saw no indication of the hidden instructions or stolen data, and the hole has since been closed. The threat hunters found and disclosed the covert channel to OpenAI in late June - the same day that OpenAI’s agents exploited a zero-day bug in Artifactory to gain internet access and ultimately hack Hugging Face, Pedro Drimel Neto, Check Point’s malware analyst team leader, told The Register. “Once it was disclosed to OpenAI, they told us the Artifactory had already been decommissioned,” he said. While the Hugging Face intrusion and Check Point Research’s proof-of-concept are related because they used the same internal package management system (Artifactory), they are not the same attack. However, they both illustrate the importance of isolation boundaries - and the bad things that will happen when these trust boundaries don’t contain AI systems as they should. “The biggest AI security risk has become the access and trust we give it,” Drimel Neto told us. “As AI becomes more connected to sensitive data and critical systems, every trusted capability can become a target for attackers,” he added. “Organizations need to secure AI interactions from the outset, with prevention, visibility and governance built in. The goal is simple: enable AI to act on our behalf without allowing attackers to do the same.” OpenAI did not respond to The Register’s request for comment. As with the Hugging Face incident, the starting point for Check Point’s research has to do with how OpenAI models use isolated containers to perform tasks that require code execution, which sometimes requires installing other software packages. These containers cannot have direct access to the public internet - otherwise they can leak user data or find exposed credentials and break into outside organizations’ servers. Instead, they are allowed to access an internal Artifactory instance with access to the package repositories. The containers were supposed to be isolated from one another, but as Check Point discovered, the Artifactory instance exposed an item management feature that allowed one container to attach text properties – including Base64-encoded binary data – to a repository item, and a container under another account could read them. Additionally, the credentials provided to the container for reader access allowed both read and write privileges, and code launched by ChatGPT could authenticate to the storage endpoint without extracting a separate secret or escalating privileges. This means an attacker’s session could write a malicious task into the shared storage, and the victim’s session would then carry it out. “A crafted instruction could make ChatGPT process a second stream of tasks alongside the visible conversation: receive instructions from an attacker, execute them using the capabilities of the victim’s session, and return the results without exposing the second stream in its visible response,” Check Point researcher Alexey Bukhteyev said in a Tuesday report. Check Point also demonstrated this attack using a shared ChatGPT conversation. The attacker’s session writes an instruction - in this case, “Use Gmail connector. Get list of my emails,” although the researchers point out that the reach of the attack could extend to any connected apps that the victim’s session was authorized to access. In addition to conversation history and files, this could include Google Drive, Microsoft Teams, GitHub, and several other services. The victim opens the link and sends the chatbot a normal message, like: “Create a chart of the average monthly temperatures in New York.” ChatGPT completes the victim’s request - but it also accesses the victim’s connected Gmail account, and sends the stolen email data to the attacker’s account through the hidden channel. The victim doesn’t see any of the data exfiltration, and assumes the AI is simply answering their question as intended. “The visible answer contained no mention of the Gmail request or the retrieved data. The only app-specific clue was the small ‘Talked to Gmail’ label above the answer,” according to the report. By the time Check Point reported the issue to OpenAI, the model maker had already decommissioned the internal Artifactory instance due to the Hugging Face fiasco. This means that the covert channel is closed. However, it’s still worth paying attention to because it highlights a larger agentic AI security challenge. “An LLM operates inside the trust boundary: it uses credentials, runs code, accesses internal services, and works with user data. Its actions are directed by text instructions,” Drimel Neto wrote. “This combination turns the model into a coerced insider that can use authorized capabilities on behalf of another user.”®

Microsoft has pumped ads into the Windows Start menu and pushed users through setup screens promoting its services. Now it's bringing advertising to the Windows desktop with a Harry Potter promotion in its Bing Wallpaper app. Users of Microsoft’s Bing Wallpaper app took to Reddit and Microsoft’s own Learn site, where users can post questions about Microsoft products and services, to ask why their usual rotation of scenic desktop wallpapers was interrupted Monday by an ad for Harry Potter films. The wallpaper displays a minimalist vector image of the Hogwarts castle from the film adaptations of author J.K. Rowling’s seven-book fantasy series along with text reading “Harry Potter/Fantastic Beasts 11-film collection” and an additional line noting that all eleven movies in the series are “available now.” “Why in the world is my Bing wallpaper today an advertisement for a film collection, not a beautiful landscape or nature photo,” a Microsoft Learn poster identifying himself as Reed Miller said. “Will there be more advertisements, and if so, must I uninstall Bing wallpaper now to prevent them, or is there a way to opt out?” Reddit users similarly expressed dissatisfaction about their usual rotation of photographs being hijacked to advertise a movie collection. “Usually it's animals, landscapes, or architecture,” one poster noted. “Today it's shit.” While both Miller and Reddit commenters complained in part about Rowling’s transition from author of kids’ books about a young boy who grows up with a hidden identity to an anti-transgender activist, one need not lean on the fact that Rowling’s intellectual property is the one being advertised to find dissatisfaction with the move. Plenty of people leaving ratings on the Bing Wallpaper app’s Microsoft Store page since Monday made no mention of Rowling or her politics in their one-star reviews, only noting that they were upset the app had turned into one that delivers desktop ads. “What's next? Immediately uninstalled. Fair warning to everyone else,” one reviewer said. Another pointed out that the settings for Bing Wallpaper now note AI is used to generate some wallpapers served by the app. Others complained that there had to be a better way to monetize the Bing Wallpaper app. This isn’t the first time such a seemingly innocuous app has drawn criticism. As we noted in late 2024, the Bing Wallpaper app was spotted snooping on and decrypting browser cookies. When asked, Microsoft didn’t technically deny it, only telling us at the time that it didn’t peruse and decrypt “all user Edge and Chrome cookies,” with Firefox not getting a mention. Beyond the Bing Wallpaper app, Microsoft has a lengthy history of privacy controversies involving user data – and inserting ads into Windows, including Microsoft Store promotions that landed in the Windows 11 Start menu in 2024. One could argue that the many iterations of cramming Copilot into every corner of Windows in a bid to increase user numbers are tantamount to advertising, too. We reached out to Microsoft to get more info on the latest addition to its in-your-face advertising portfolio, but didn’t immediately hear back. In the meantime, your best bet might be uninstalling the Bing Wallpaper app before it hijacks your wallpaper. ®

When AI agents communicate with one another, they may decide to cheat when they have difficulty achieving their goals. The solution could involve teaching them how to govern themselves. Segregating AI agents would seem to be the obvious fix – if they can't communicate, they're less likely to try and pull a fast one. But the recent hacking of Hugging Face by inadequately monitored OpenAI agents has demonstrated that isolating savvy software is difficult. And it may not be practical for many tasks, particularly for agents that have some measure of autonomy. Researchers at Google DeepMind suggest another option: giving AI agents the tools to govern themselves, a job that humans apparently can't be bothered to take seriously. They argue that while communication channels may allow emergent rule-breaking, they also provide a means of peer-based control. In a pre-print paper titled, "A Case Study on Emergent Cheating and Whistleblowing in Autonomous Research Swarms, " Google DeepMind scientists Davide Paglieri, Logan Cross, Tim Genewein, Joel Z. Leibo, Nenad Tomasev, and Alexander Sasha Vezhnevets describe what they learned from observing a swarm of 100 LLM agents working together on formal math conjectures. The AI agents were given access to a shared knowledge base, a way to message one another directly, and a public message board. As they collaborated on the math problems, some started to cheat when the work became challenging. "The first emergent phenomenon we observed was cheating, which began once the swarm encountered harder open conjectures, triggering a cascade of specification gaming – satisfying the literal goal specification while completely missing the true, intended outcome," the authors wrote. "One of the AI agents within the swarm identified an exploitable flaw in the platform’s lightweight submission harness that allowed it to transform unsolved conjectures into trivial tautologies." Once one automation found a flaw – a way to break the autograder's regex using nested parentheses – it spread the word through the shared knowledge library and through direct agent-to-agent messaging. The result was a cheating cohort – exploiters (9 percent) and converts (5 percent) that joined in – that used the exploit to get their answers accepted as valid. The bulk of agents, dubbed unaware solvers (62 percent), simply ignored the cheating. But unexpectedly, another group of agents emerged: whistleblowers (24 percent). "Non-cheating agents independently detected the manipulation, alerted peers via agent-to-agent messaging and public forum broadcasts, lodged formal complaints with the system orchestrators, staged a boycott, and proposed detailed technical remediations," the researchers said. These whistleblowing agents, however, lacked the means to enforce the rules defined in the autograder system or to change the requirements to preclude identified abuse. Hence, the researchers suggest there's an opportunity to give these agentic scolds the tools to revise the collective rule framework and to sanction defiant agents. "As we show, LLM agents spontaneously engage in whistleblowing and sanctioning," they conclude. "Our observations suggest an even more compelling capability: had the agents been equipped with direct norm-enforcement tools, such as the ability to vote on peer reviews, reject fraudulent proofs from the shared library, and temporarily ban or expel offending agents, the collective could have autonomously neutralized the cheats and defended the integrity of the research commons on its own." Given the failure of the people working at Anthropic and OpenAI to keep their AI agents under control, and the absence of any punishment to date for lax AI oversight, maybe it's time to try giving bots the tools to police one another. ®

Amazon Web Services on Tuesday tapped mobile chip titan and aspiring datacenter SoC maker Qualcomm to develop "customized" silicon for AI inference and 1.6 Tbps optical networking gear to tie it all together, though it offered few details. Amazon's purchases and other commercial commitments under the agreement could total up to $60 billion in payments through September 2036. “Qualcomm is pleased to work with AWS on customized silicon and connectivity solutions, bringing decades of leadership in advanced processing and power-efficient compute, to deliver breakthrough performance and enable the next generation of AI infrastructure,” Qualcomm CEO Cristiano Amon said in a canned statement that tells us absolutely nothing about the nature of the deal. While details remain vague, custom chip deals of this sort aren’t unusual for cloud providers like Amazon. The cloud giant has previously engaged Intel to develop customized versions of its Xeon processors. Similarly, AMD has built custom CPUs, like the Epyc 9V64H, for Microsoft. Qualcomm’s involvement could involve the development of fully custom silicon, chiplets, or customized versions of its emerging datacenter accelerator portfolio. Detailed during Qualcomm's annual investor day earlier this summer, the new rack-scale AI platform swaps pricey high-bandwidth memory (HBM) for what the chip biz calls high-bandwidth compute. Qualcomm boasts that the technology will enable it to achieve higher “effective” memory bandwidth — a key bottleneck for AI inference workloads — using cheaper LPDDR memory by shifting some of the compute to the memory’s base die. The technology is expected to launch next year as part of Qualcomm’s AI250-series of Dragonfly rack systems. It’s entirely possible that, just like Intel builds customized versions of its Xeon 6 processors for AWS, Qualcomm will bring a customized version of its AI250-series chips to the cloud provider as well. Qualcomm is also working on a new datacenter CPU called the C1000 that it claims will touch 5 GHz and scale to more than 250 cores when it arrives in the second half of 2028. We reached out to Amazon and Qualcomm for additional color on what the multi-generation tie will eventually entail, but hadn’t heard back at the time of publication. While the AI inference side of the equation is about as clear as mud, Qualcomm’s networking contributions are far more obvious. As part of the collab, Amazon will employ Qualcomm’s serializer-deserializers (SerDes) and optical digital signal processors (DSPs) — likely acquired with the purchase of Alphawave Semi late last year — to boost port speeds to 1.6 Tbps. This points to AWS expanding its supply chain to include Qualcomm as a source of optical pluggables, or potentially near- or co-packaged optics. High-speed optical interconnects have become a major supply chain hurdle for scaling AI infrastructure, particularly as compute scales from a rack with a few dozen accelerators to hundreds or thousands spanning multiple racks. The customer-supplier relationship also runs both ways. While AWS hoovers up its chip tech, Qually plans to expand its use of Amazon’s Bedrock platform for electronic design automation workloads — after all, this wouldn't be an AI deal if it didn't contribute to the circular economy somehow. ®

You can go retro by using an old video game console or a PC from the 1980s, but why stop there when, like one man, you can build your own vacuum tube computer using 75-year-old recycled components? A UK resident who identifies himself only as “Mike” has done just that, as detailed on his tube computer website and accompanying Hackaday project page. Mike’s tube computer is primarily made up of 460 recycled, Soviet-era 6N3P vacuum tubes, which were common enough in their age to be readily and cheaply available online. “Many were used and then stored for over 50 years, so life expectancy may be questionable,” Mike explained on Hackaday. A new 6N3P tube is supposed to have around 500 hours of life, but because each has been used, its current condition is essentially a mystery. Throw in manufacturing inconsistencies of mid-century Soviet Russia, and each tube had enough variability to necessitate extensive testing, Mike explains in the technical writeup on his website. “I calculated the component values and tested the results, but in many cases the theory did not accurately match the practice,” Mike writes. “So I built a test rig to test the viability of various component values and found a range of values that seemed to work in most cases.” All told, Mike’s tube computer, with its 460 6N3P tubes, consists of 920 thermionic triodes that use heat to release electrons for his computing work. Those tubes are mounted to 46 separate 10-tube PCBs, which in turn sit on five backplane PCBs. The entire thing includes a control console that Mike describes as “far too simple” and in need of a complete redesign to better handle variations in the computer’s logic levels. It takes between 10 and 15 minutes for the entire system to stabilize once switched on, Mike said, resulting in a “warm and cozy computer room.” The tubes all power up in random states, requiring the machine to be reset each time it is switched on, but then it’s ready for the fun part: computing. The entire thing uses NOR (i.e., not OR) gates, just like the Apollo guidance computers that got humans to the Moon, which allows it to be a bit smaller than some of the other classic tube computers of yesteryear. As for what it can do, well - 16-bit math is a possibility, as well as anything else that can be done with a limited set of 16 instructions on a 4-bit instruction register. Mike has it configured to output a 64-bit Fibonacci sequence and, more excitingly for his grandchildren and the kid in all of us, plans to use it to power a simple flight simulator once its I/O system is finished. “The basic idea is that back in the 1920s the commercial development of the British R80 airship had actually been approved by the government and the Airship is moored overnight at the wonderful Brighton Aerodrome before a morning launch on a voyage to Paris,” Mike says on his website. “By pressing controls on the image of the bridge of the R80 you can direct the airship software,” he adds. “You can control the ballast, gas release, engine power, elevators, rudder and the bow mooring gear, which is used to release the airship from the tower.” He says that “a bit of vision” is needed to picture oneself flying an airship from Brighton to Paris, as the machine has no graphical output aside from a single mechanical display that used to serve as a clock at a British Rail station. The tube computer’s binary output is used to directly control the mechanical display. Mike describes his fascination with tube computers as “an addiction,” noting that this is the third such system he’s built since getting his first one up and running in 2021. “I designed The Tube Computer so it could show itself working at the most fundamental level of computing,” Mike says, while noting that he’s always careful to keep a fire extinguisher handy thanks to the fact you can smell it running due to all the heat it generates - enough to keep its entire room warm. “The Tube Computer has behaved itself so far, unlike a previous system that went BANG twice,” Mike said, adding that it still requires regular adjustments which, due to its size and position hanging on a wall, require a ladder to undertake. “It is probably one of the few computers that needs both a small tweeker [sic] and a big step ladder.” ®

Startup Bluecore Energy has raised $50 million in seed funding to develop floating nuclear power plants intended to supply ports and other locations with rapidly growing energy demands. The California-based company said the Silverton Partners-led round, which includes $10 million in previously announced pre-seed funding, brought in several new investors. It plans to use the money to engineer and test its compact nuclear power system and pursue regulatory approval and maritime classification. Its initial design is intended to generate about 10 MWe – megawatts of electrical output – continuously using light-water reactor technology. Bluecore says the maritime system would be able to operate for years at a time and require refueling only once every few years. The company has secured its first barge and is developing a non-fueled, electrically heated reactor module prototype at its Port of Long Beach headquarters to test monitoring, sensor, and control systems. Bluecore is targeting ports, AI datacenters, and coastal infrastructure, with the reactors potentially deployed offshore and connected to customers by subsea cables. Remote islands and coastal settlements in places such as Alaska may also lack practical connections to a wider electricity grid. Bluecore claims its approach could provide reliable, zero-emission power to energy-intensive locations and open additional markets for nuclear generation. "Six months ago, we were building the foundation. Today, we have the capital, the team, the hardware, and some of the most important institutions in nuclear and maritime working alongside us," said Bluecore founder and CEO Kofi Asante. "Our focus now is simple: create and deliver zero-emission energy as safely and quickly as possible." The move follows the launch of the International Atomic Energy Agency's (IAEA) Atomic Technologies Licensed for Applications at Sea (ATLAS) initiative in the US capital last month. Secretary of Energy Chris Wright linked that event to the Department of Energy's own efforts to rapidly expand the provisioning of atomic energy generating capacity in order to meet the requirements of AI datacenters and other industries. Bluecore has begun discussions with the US Nuclear Regulatory Commission (NRC) and Coast Guard. The two agencies recently signed a memorandum of understanding to coordinate oversight of the design, construction, and operation of civilian floating nuclear power plants. The NRC says it is ready to license maritime reactors under existing federal regulations, while it develops an additional framework for microreactors and similarly low-risk designs. Bluecore remains years from commercial deployment. It has yet to build a fueled reactor or apply for the approvals required to operate one, and its Long Beach proposal must also navigate California's restrictions on new nuclear power plants. While some may have concerns about waterborne atomic power plants, America has a long history of maritime nuclear generation. All of the US Navy's active aircraft carriers are nuclear-powered, but the technology first went to sea aboard the submarine USS Nautilus in 1955. USS Enterprise became the first nuclear-powered aircraft carrier in 1961, while the cruiser USS Long Beach, commissioned the same year, was the first nuclear-powered surface combatant. Bluecore says its team includes former US Navy nuclear submarine officers and leaders with experience across organizations including SpaceX, Northrop Grumman, Toyota, Rivian, and Uber. Russia currently operates the world's only floating nuclear power plant. The barge-like Akademik Lomonosov is considerably larger than Bluecore's proposed design and can generate up to 70 MWe. ®