SpaceX is aiming for a September 22 launch of its monstrous Starship rocket on a 10-hour mission to deliver 26 Starlink V3 satellites into orbit. The flight, set to launch at 1215 UTC if all goes to plan and the authorities give the green light, will be the first time Starship has gone into orbit, and the mission is planned to fly at an altitude of approximately 275 km. The mission aims for six orbits, followed by a deorbit burn using a single Raptor engine while. The engineers are then planning a splashdown in the Pacific Ocean west of Chile. The Super Heavy Booster will be ditched in the Gulf of Mexico, hopefully under better control this time. On flight 13, the booster made a "hard splashdown," as SpaceX delicately put it, after only eight of the planned 13 engines reignited. The booster used all 33 engines for its boostback burn, but according to SpaceX, "the three center engines showed signs of ice clogging which triggered an early end to the maneuver." In true iterative style, SpaceX tweaked the hardware to improve filtering and updated the software to "enhance relight reliability." SpaceX has also changed Starship's heatshield, including adding curved tiles to reduce heating in tile gaps and "additional retention mechanisms added to tiles in areas deemed to be at highest risk of falling off during ascent." It is also reusing a pair of tiles recovered from the previous flight test. If Starship is to become truly reusable, these tiles will need to fly over and over again. Getting to orbit is by no means guaranteed. The SpaceX team will only command a final burn to enter orbit if there is sufficient hardware redundancy to perform a deorbit burn when the time comes. After all, nobody wants something the size of Starship making an uncontrolled reentry after its orbit eventually decays. Starship has had an eventful test program. Some missions were spectacularly successful, and several boosters returned to be caught by arms on the launch tower. Others were impressive failures. The vehicle is critical to SpaceX's future; it can loft far greater payloads than the company's workhorse Falcon 9 and is an essential part of NASA's Artemis program. For Artemis III, planned for 2027, SpaceX is expected to launch a Starship equipped with the docking apparatus planned for the lunar variant. Reaching orbit is therefore a critical step. ®

Amazon Web Services (AWS) says it is unable to restore access to resources and data in some of its Availability Zones in the Middle East after datacenters were damaged during the US war with Iran. In an update to its AWS Health Dashboard, the cloud giant confirmed that anything hosted exclusively in its Bahrain Region (me-south-1) remains inaccessible. "The damage to our infrastructure spanned multiple Availability Zones and exceeded what our regional and multi-AZ services are designed to withstand," the update says. "After a thorough assessment, we have determined that we are unable to restore access to the resources and data hosted exclusively in this Region." After the first Bahrain Availability Zone was damaged in March, AWS advised customers to migrate their workloads to other Regions. The company said most did so before further attacks disrupted a second Availability Zone in April and rendered the entire Region unavailable. AWS has reached a similar conclusion about one Availability Zone in the United Arab Emirates (UAE), saying it cannot re-establish access to the resources and data hosted there. The affected Zone, mec1-az2, is one of three in AWS's UAE Region. Two AWS facilities in the country were hit by drones back in March as Iran retaliated against US-Israeli strikes on its territory. AWS says it continues to work on recovering resources in the Region, including some hosted in the other affected Availability Zones (mec1-az1 and mec1-az3). It adds that "most customers" have been able to restart operations in other AWS Regions by restoring from backups or copying any data that remained accessible. The company says its support teams remain available to help customers move their applications to alternate Regions. The Register asked AWS to comment, but the company declined to add to the information on its Health Dashboard. In AWS terminology, a Region is a geographic area containing multiple isolated Availability Zones. Each Zone consists of one or more discrete datacenters with independent power, cooling, and network connectivity. AWS recommends distributing applications and resources across multiple Availability Zones to withstand the loss of a single location. That approach cannot protect against a Region-wide failure, however. Workloads that must survive such an event require a disaster recovery plan involving another Region. AWS and several other tech companies advised customers after the first strikes to recover whatever resources they could and replicate them in alternate Regions, ideally in Europe, rather than wait for services in the Middle East to return. That advice appears to have been well-founded. AWS said at the time: "Customers should enact their disaster recovery plans, recover from remote backups stored in other Regions, and update their applications to direct traffic away from the affected Regions." ®

Spain’s data protection agency (AEPD) has reported the country’s first-ever personal data breach caused by the actions of an autonomous AI agent. Francisco Pérez Bes, president and deputy of the AEPD, said in a Monday blog post that an individual deployed an AI agent that used a “known large language model (LLM)” to carry out the attack on an organization. The agent scanned “generic files” before accessing the organization’s system, then ran vulnerability scans to find flaws that would give it read/write access to files containing personal data and invoices. Pérez Bes did not name the LLM used to support the attack, but said whoever was behind it used the agent to “successfully chain together different phases of the attack.” This demonstrates that AI-supported attacks are no longer theoretical, he added, and called on organizations to embrace defense tools that are capable of keeping pace with the speed at which agentic attacks can be executed. “Human supervision remains essential, but it must be supported by detection, containment, and response mechanisms capable of operating quickly enough,” said Pérez Bes (machine-translated). “The arrival of AI agents in the offensive arena should prompt an immediate review of security and data protection models. “Data protection officers, managers, and delegates must prepare for a scenario in which the speed of attacks will increase, but in which the same fundamentals will continue to be crucial: Understanding the processing activities, minimizing data, limiting access, correcting vulnerabilities, controlling suppliers, and being prepared to respond.” The Register asked AEPD for more information. Spain’s first AI agent attack comes as the AEPD recently recorded its busiest year for data protection complaints. According to its most recent annual report, covering 2025, the agency received 30,931 complaints – the most in its history – representing a 64 percent increase compared to the year before. And although Spain is only now getting its first taste of a security mishap caused by a naughty agent, cases involving the foremost US AI houses are already heavily documented. OpenAI’s claim in July that its agents escaped a sandbox and started attacking Hugging Face kickstarted something of a battle between it and rival Anthropic over whose agents could take the most liberties with their security. Both companies have reported several instances of their agents going rogue, escaping "secure" environments and going walkies across the internet to attack unwitting organizations. OpenAI has been circumspect about the true scale of its rogue agents’ damage, as third-party reporting showed more websites than it was letting on were taken over. Similarly, Anthropic has said that its AI agents had, in four cases now, accessed third-party systems in attacks that, if carried out by a human, could see them convicted under computer laws. ®

Logitech has released the MX Keypad, a developer-focused keypad derived from its MX Creative Console, and something that triggers a distinct sense of déjà vu. Do you remember Stack Overflow's "The Key" Joke? Three keys to copy and paste code from the developer's former best friend (until AI tools began to eat its lunch). Logitech took the keypad from its MX Creative Console product and made something similar to Stack Overflow's joke-that-became-a-thing, though considerably more extensible. The company sent us one to look at, and it's an interesting, if perhaps slightly expensive, bit of kit. At its heart, the MX Keypad is a set of nine keys that can trigger commands. Out of the box, it all looks a bit bland – there are monochrome images on the keys for playback, volume, screenshot taking, and even something to bring up an emoji panel when attached to a Windows device. However, things get a good deal more colorful and interesting when the device is connected to the Logi Options+ application. Then the device moves from being a curiosity to something that threatens to be indispensable, depending on your workflow. Using Word to knock out some writing? The MX Keypad will show a bunch of commonly used commands, or you can customize it to suit your workflow. How about Visual Studio Code? It's the same again – the keys light up with some default commands that you can easily make your own. According to Logitech, the device is a "Multi-App AI Control Center for Developers" and was developed in conjunction with keyboard jockeys. We'd call it a surprisingly useful bit of kit which has rapidly become part of our workflow, both for productivity and when diving into VS Code for some source wrangling. There are, of course, downsides. Although the price is relatively competitive, it is a little on the high side for what is ultimately a smart macro keypad. It also requires a USB-C connection, which we can understand given the power needs of those bright, backlit, color LCDs, but it would have been nice if Logitech could have come up with something that didn't need a cable. And, of course, you need Logitech's software to get the best out of it. In the box is the keypad itself, a USB-C cable, and a stand. We preferred to position it on the desk next to the keyboard for ease of access – the device has a slightly angled base, so it works well in that placement, but the inclusion of a stand is a nice touch. There's also an SDK if engineers (or their agents) wish to get their hands dirty, and a marketplace where plugins for the likes of Claude Code and OpenAI Codex can be found. According to Logitech, these give developers "real-time agent status tracking and CLI control right on the physical keys." Assuming you can make it work. We found the Logi Options+ software a little unintuitive and recommend some perseverance to get the best out of it, along with a little caution, since the built-in designer can create some deeply wince-inducing key designs. Ultimately, the MX Keypad represents some excellent repurposing by Logitech. The original MX Creative Console was useful for creative types, but its jog wheel was arguably of little use to developers. The touch panel, with its customizable keys, is a much more practical proposition; by selling it on its own, Logitech has cut the price to £89.99 ($99.99). It comes in light or dark gray and has a pleasingly quality feel. Having to plug it in is annoying, given the wireless nature of many other peripherals, and you do need to get to grips with the Logitech Options+ application to get the best out of it, but once it forms part of your workflow, it becomes a hard habit to break. ®

The Ministry of Justice (MoJ) has apologized after court staff accessed documents relating to victims and survivors of the 2024 Southport murders without authorization. For a limited number of people, the material accessed included sensitive personal data assessed as likely to pose a high risk to their rights and freedoms. There is no evidence that the information was shared with third parties. "We are appalled that this happened and recognise the distress it will have caused victims, survivors, and their families," an MoJ spokesperson said. "We apologise to those affected – unauthorized access to court files is completely unacceptable. "This is now being investigated urgently, and the prime minister has asked the Lord Chancellor to oversee this. "All wrongdoing will be met with extremely firm action." The Register asked the MoJ how many staff accessed the files, whether they remained employed, and what their reasons may have been. It did not address those questions. Those affected include members of victims' and survivors' families, all of whom are being contacted directly. The MoJ did not disclose how many people were involved. HM Prison and Probation Service and HM Courts and Tribunals Service are investigating the matter. The Information Commissioner's Office has also been informed. The MoJ breach is the latest in a series of incidents involving inappropriate access to sensitive records connected to the attack. Separately, North West Ambulance Service investigated potentially inappropriate access by some of its staff to records of patients in the Southport attacks. And nearly 50 staff were found to have inappropriately accessed the medical records of some victims treated at Aintree University Hospital, near the place of Axel Rudakubana's attacks. Rudakubana, who was 17 at the time and has since been admitted to a psychiatric hospital, attacked a Taylor Swift-themed dance class in Southport, England, on July 29, 2024, killing three children and injuring eight other children and two adults. False claims about the attack online prompted violent, racially charged riots across the UK. Police made 1,511 arrests in the weeks that followed and brought 960 charges. Rudakubana was sentenced to life imprisonment with a minimum term of 52 years. ®

UPDATED An hours-long Salesforce outage on Wednesday hit customers with severe delays, intermittent errors, and inability to access some services, according to the SaaS vendor's status page. The outage hit hundreds of instances throughout the world, including the USA, Japan, India, the UK, France, and Germany. The service disruption was first reported around 0930 BST (0830 UTC) on September 16. An update at 1219 BST (1056 UTC) said the company had validated and rolled out a fix, and customers were seeing service return to normal. However, as of 1500 BST (1400 UTC), the company said that the fix had not completely restored all instances of the problem, and it had "received reports of scheduled jobs not running as expected for some customers who are now able to access the service." A previous update at 1010 BST (0910 UTC) by the CRM giant said its investigation indicated that "requests are stalling while waiting on a response from an internal login service, which is using up available server resources." The vendor added: "Customers continue to experience severe delays, intermittent errors, and inability to access some services. This also affects support case creation." Salesforce's customer list is a roll-call of global giants from all industries, including include Amazon, Walmart, Coca-Cola, Toyota, and IBM. The outage came at a bad time for the SaaS company. Yesterday, it kicked off its annual conference, Dreamforce, in San Francisco. More than 40,000 people are expected to attend in person, with in excess of 200,000 people registered to attend online. Updates to the situation, which is still ongoing as of 1725 UTC, are below. ® Updated at 2020 BTS (1920 UTC) Salesforce says that the incident has now been resolved. Updated at 1722 BTS (1622 UTC): Salesforce says that the impact radius is narrower than they originally thought. They will continue to update impacted instances. Updated at 1650 BST (1550 UTC): Salesforce has completed the next phase of its rollout and narrowed the remaining disruption to a subset of Hyperforce instances. Its first-party environments are unaffected. Mitigation is in place across nearly all affected instances, although the company continues targeted restarts and manual recovery work where the automated fix has not fully resolved the problem. Some customers could still experience delays, errors, and problems accessing services or creating support cases. Updated at 1522 BST (1422 UTC): Salesforce said it was manually restarting instances where the automated fix hadn't fully resolved the issue, and that it was still receiving reports of scheduled jobs not running as expected for some customers who were able to access the service. It also said that the "impact radius" of the problem was "narrower than initially understood." Updated at 1429 BST (1329 UTC): Salesforce said the rollout had not completed successfully on some instances, leaving customers exposed to the original problem. The company is reapplying the fix and warns that some users might need to clear cached data or restart their sessions to recover fully. Updated at 1407 BST (1307 UTC): Salesforce said most instances were recovering as the rollout progressed. Some customers may continue to encounter errors until their instance is manually restarted, and the company has yet to confirm when the incident would be fully resolved. Updated at 1327 BST (1227 UTC): Salesforce said services were returning to normal in regions where the rollout had been completed. The company continues to monitor the deployment while working on a permanent fix. Updated at 1300 BST (1200 UTC): Salesforce said the fix was being deployed region by region and that service was improving, although it had no confirmed completion time. GovCloud customers were no longer affected, but others could continue to experience severe delays, intermittent errors, and problems accessing some services or creating support cases. Updated at 1227 BST (1127 UTC): Salesforce said the fix was being rolled out "fleetwide" and that customers were seeing service return to normal. Updated at 1200 BST (1100 UTC): Salesforce said the fix had been successfully validated on a test instance and would now be deployed across all affected instances. Updated at 1142 BST (1042 UTC): Salesforce said increased load on a core system component had limited its capacity to process requests. The company is testing a fix on one instance and plans to deploy it across its fleet once validated. Updated at 1109 BST (1009 UTC) on September 16, 2026, to add: Salesforce said in its latest update that it is "no longer pursuing restarts as a path to remediation. Customers continue to experience severe delays, intermittent errors, and inability to access some services and support case creation."

BT Email customers say they have been bombarded with hundreds – and in one case more than 1,000 – unsolicited password reset messages, leaving the telco investigating what is causing the deluge. The reports began appearing on BT's community forums over the weekend, with customers describing sudden bursts of PIN messages arriving within minutes despite them making no attempt to reset their passwords. One customer said they received between 50 and 70 PIN texts "in the space [of] a few minutes," while a second reported more than 100. A third said the number had passed 500, and another claimed to have received more than 1,000. One BT customer who contacted The Register separately said they had received around 300 password reset emails over the previous 24 hours, typically arriving in bursts of around 50 within a minute before stopping for several hours. BT has acknowledged the problem but has yet to explain what is behind it. "We are aware that some customers have received a large number of password reset pin messages and we are actively investigating the cause of this," a BT moderator said on the company's community forum. "If you have received multiple messages like this, please ignore them and remain vigilant for any unusual activity with further messages or unexpected calls." BT added that customers' email accounts are secure and that those receiving the messages do not need to take any action. At least one customer, however, claims someone accessed their account during the PIN barrage. The user said they lost access to both their email and BT ID overnight, but managed to request a PIN before whoever was accessing the account could change the telephone number attached to it. That account has not been independently verified, and it isn't clear whether the apparent takeover was connected to the flood of PIN messages. The cause of the wider problem remains a mystery. It isn’t yet clear whether it’s the result of someone deliberately hammering a password recovery mechanism, a fault within BT's systems, or something else entirely. The Register has asked BT how many customers are affected, whether the messages are being generated by external password-reset attempts, and whether it has found any evidence of malicious activity. We've also asked the telco what rate limiting it applies to password reset requests, given reports of hundreds of PINs arriving within minutes, and whether both email and SMS messages are affected. BT says it's investigating. In the meantime, some customers are getting enough PINs to start their own lottery. ®

The UK's Advanced Research and Invention Agency (ARIA) is funding projects aimed at delivering connectivity using high-flying aircraft, including technology to beam power from the ground to keep them in the air. Britain's science and technology body, inspired by America's DARPA but on a more modest budget, says it is backing 18 projects to develop aircraft capable of operating reliably and cost-effectively in the stratosphere, as part of a program known as Enduring Atmospheric Platforms. Success will be measured by whether any aircraft can continuously supply 300 W of power to a communications payload for an entire week, while holding station above the UK. The intent is to create an alternative to orbital satellites for providing connectivity to underserved parts of the UK and elsewhere in the world. These are known as high-altitude pseudo-satellites (HAPS). The agency says that satellites have been the default way to reach users or communities that are not well served by terrestrial networks, but their distance and orbits impose physical limits on what they can deliver. ARIA is allocating £70 million ($94 million) in funding over three and a half years for the projects, which will tackle the power and reliability challenges of sustained operations above the UK. The projects are split into three technical areas, with TA1 set to cover enabling technologies, TA2 comprising system integration and testing, and TA3 focused on deployment and communications. ARIA highlighted a project called Carousel from aerospace startup Menapia. This plans to produce a constellation of high-altitude multi-rotor aircraft, capable of routine operations at altitudes up to 20 km (12 miles), with the intention of forming a persistent "canopy" over the UK. Another is Enduring Drone Networks, being developed by VFP Aerospace, which is aiming for persistent presence via a relay team of units that will cycle between their service altitude and ground stations for recharging. It says it is targeting a production cost of £50,000 ($67,000) per drone. In the enabling technologies category, a firm called Hypanode Ltd is working on a fuel cell power system to deliver 20 times more energy than current battery technology. Scalable Laser Ltd is developing laser-based power beaming technology, pairing high-power laser diodes with advanced photovoltaic receivers, while Space Solar Engineering Ltd is looking at radio frequency power beaming to deliver continuous power to a HAPS aircraft. A more exotic-sounding technology is Stratospheric Navigation and Gravity (STRAT-NAV) from the University of Bath. This seeks to extend the endurance of HAPS craft by allowing them to predict and make use of atmospheric gravity waves – caused when air is forced upward by weather fronts or hills, and gravity pulls it back down, creating a ripple. "This research expands the UK's leadership in high-altitude platforms to become the place where this industry is designed, built, and scaled," says Rico Chandra, ARIA's program director for Enduring Atmospheric Platforms. "Getting there means backing bold ideas, from fixed-wing solar aircraft to designs nobody predicted, including aircraft that fly on spinning wings instead of propellers." ARIA funds teams of scientists and engineers to pursue cutting-edge research. It is a non-departmental public body sponsored by the Department for Business, Innovation, Science and Trade (BIST). It was until recently chaired by entrepreneur Matt Clifford, who stepped down after taking up a role with US AI firm Anthropic. The agency was also controversially spared being asked to make savings leading up to the UK government's Spending Review last year. ®

Unpredictable token bills haunted the first half of 2026 as AI experimenters pushed deeper into the tech and often found useful results could only be achieved after surprising and unwelcome expenditure. We've gone from tokenmaxxers to tokenmisers, counting every character we send to AI services. That changed the way we used AI, focusing users on the ticking meter and unable to cast their eyes up and explore the possible. Those blinkers fell away in the middle of August when Alibaba delivered Qwen3.8-27B, byte-for-byte probably the most powerful language model ever created – and also capable of running on a high-spec laptop. The community of “local AI” enthusiasts noticed and have spent the intervening weeks whittling Qwen down to fit on an ever-widening array of devices, including – surprisingly – my M4 MacBook Air. Their work means millions of similar devices can now access near-frontier-level artificial intelligence from free software. The gap between my MacBook Air and a smartphone isn’t vast. As local AI developers continue whittling Qwen down to size, a few hundred million smartphones will also be capable of running top-tier AI. While that hasn't happened yet, it won't be more than six months away. At the rate things are unfolding, probably closer to six weeks. Suddenly, "good enough" AI isn't something that we will have to rent. We can have it, at no cost, on our desktops, our kitchen tables, our classrooms and in our pockets. That’s great, but also scary because we've been caught unprepared by AI effectively becoming free. Nobody yet knows what we will do with free AI, or if we are ready for it. On the positive side of the ledger, we stand to benefit from a growing “cognition surplus” that brings more thinking capacity into the home or office. Need to wrangle your way through a baroque accounting system? Give it to the AI. Sorting through twenty years of smartphone snaps? Again, let the AI do the heavy lifting. How about a personal newspaper gathered while you sleep, and presented on your tablet at the breakfast table? It's all doable now. All of that feels very much like something I remember clearly from the early 1980s, when the promise of the “home computer” was managing recipes. We knew the device would find a use, but it wasn't until the one-two punch of GUIs and the Web that most people had any real use for a home computer. That's exactly how this moment feels. Sure, managing smartphone photos will be helpful for millions, but the biggest win will be on the darker side of the ledger – where every bad actor – from script kiddies with a grudge through to the worst of the worst – now have equal access to powerful AI that will go with them everywhere. They’ll use it to lie, cheat and vex, at a speed and on a scale never before seen. To deal with that, we'll fight fire with fire. The most important job our AIs will be tasked with will be keeping us as safe as possible from other AIs. We’ll only be aware of this when it fails. But “good enough” AI and “intelligence everywhere” will enable resource pools and defensive techniques that our adversaries could never harness for themselves, so we stand a solid chance. ®

When Microsoft delivered over 970 patches last week, many saw a nightmare for beleaguered security staff. Gartner research vice president Craig Lawson thinks infosec workers might soon see sunlit uplands as their workloads ease. Lawson outlined that scenario at Gartner’s IT Symposium in Australia today and explained it by theorising that the increased volume of vulnerability discoveries made possible by Anthropic’s Mythos and other bug-hunting AI might be getting close to finding most of the flaws in established codebases. “We've never had a situation where massive codebases have been audited to that level before,” he told The Register, and offered the recent series of CVEs found in OpenBSD – which has historically been an unusually secure and stable OS – as evidence that AI bug-hunters are cleaning up. “Think about how much technical debt has been retired in products just in the last six months,” he said. Lawson pointed to the fact security vendors, who in theory know what it takes to create secure products, are also using AI to find flaws in their wares. Those discoveries, he suggested, again indicate AI is taking out potential avenues for zero-day attacks. The high number of CVEs reported in 2026 is a positive signal. Lawson thinks Mythos and its ilk may also create an invisible signal as vendors use the AI to detect more bugs in their future releases. He therefore thinks that 2027 might see CVE numbers fall as vendors finish cleaning up old codebases, and because they use AI to more thoroughly test their next releases. “2027 could be the first year we see a net drop, maybe not in aggregate vulnerabilities, but definitely in severity of flaws,” he told The Register. He thinks AI will also make defenders happy by giving them better tools. Today, he said, a red-teaming exercise is an infrequent and costly event that usually involves hiring an external provider. AI bug-hunters could mean organizations can effectively run a red team every day. And if a red team exercise produces tickets that need solving, he thinks AI will help analysts to identify fixes more quickly. “What if I could spend three minutes going to Gemini and saying ‘Write syntax for an F5 IRule’ that becomes a virtual patch? Everyone can do threat intelligence, enrichment, some of those harder tasks.” When infosec staff make those fixes, Lawson wants organizations to celebrate the impact of their work. Today, Lawson said, security operations centers measure staff by the number of tickets they process and close. He thinks a better approach is to celebrate the fact that cyber-defenders kept a hospital open or stopped a ransomware raid. ®

If one sign of maturity is making better choices, then Java has grown up a fair bit in its version 27 release Tuesday. “JDK 27 is the release where Java stops asking you to opt in to good defaults and simply turns them on,” wrote backend engineer Arvind Kumar, on his education site Codefarm. Better default settings may not be groundbreaking, but they do make life easier for developers because they mean coders have fewer things to change. Nor should users have to grapple with new features, particularly if they are not of interest. Java 27 succeeds on both counts. Built from the now-completed Java 27 specification, JDK 27 is a short‑term release that won't get long-term support. It includes four new features, or JEPs (JDK Enhancement Proposals) that are ready for production, none of which, Kumar points out, require any immediate changes in developers' code. The release includes another five new features offered as previews for further incubation. Headers get compact In Java 27, compact headers are now available by default (JEP 534). Up until now, the developer had to set them with a flag. Now, no additional action is required. Traditionally, a Java object header takes up 96 bits in a memory heap, much of which is rarely used. With Java 27, the default is a 64-bit header. Turns out the header had too much empty space. Four bytes doesn’t sound like a change to worry over, but across millions of live objects it can lead to considerable memory savings. A run on the SPECjbb2015 Java benchmark standard found that toggling to a 64-bit header resulted in 22% less heap space and 8% less CPU time. The smaller headers appear to work in production. Amazon already runs hundreds of production services with the diminutive compact object headers. SAP has already switched to compact object headers as the default for the company’s OpenJDK fork, SapMachine. Even in its smaller size, this new layout reserves some space for future projects, such as 4 bits for Project Valhalla which aims to bring value objects to Java. Java's overseers don’t want to toss the 96-bit header layout, at least for the time being. But now developers must raise a flag if they want to use it. Quantum-proofing cryptography Looking ahead to the arrival of quantum computing, Java maintainers have implemented “post-quantum hybrid key exchange,” based on TLS 1.3 (JEP 527). Java now offers three new “hybrid” ways of exchanging keys between Java clients, servers and external TLS-compliant endpoints. Each of these approaches, which are options alongside the existing key exchange protocols, combines a traditional Ephemeral Elliptic-Curve Diffie-Hellman (ECDHE) algorithm with the NIST ML-KEM post‑quantum key exchange algorithm. The idea is to prepare now for future quantum attacks when the keys to traditional key exchanges could be easily cracked with quantum computing. Malicious attackers are already saving encrypted sessions with a "harvest now, decrypt later" mindset. Taking out the trash, in a good way JDK 27 also tidies up its garbage collectors, the tools that quietly identify and reclaim unused memory. It switched out the default trash bin for constrained environments – Serial – to the overall default garbage collector, called Garbage-First or G1. Garbage collectors relieve developers from manually performing this task and potentially messing it up (usually). But it comes with a cost. Aggressive memory management adds overhead. So different garbage collectors have been built to meet different performance requirements regarding throughput, latency, footprint, and startup time. Recent performance improvements in G1 effectively rendered Serial GC unnecessary, Java's bosses argue. Serial was a single-threaded app built to run on a single CPU, making it well-suited to embedded and other resource-limited environs (especially with clients with less than 1792MB of memory). Thanks to optimization over the years, “G1's performance is now sufficient to replace Serial in all situations in which the JVM would previously have selected Serial,” the maintainers write in JEP 523. With this change, embedded developers will no longer have to add explicit flags to the command line during startup. Even better, it will correct cases where G1 should have been assigned in the first place, such as for containerized apps. “A chunk of containerized Java apps has been running on [Serial] all this time, even though no one consciously made that decision,” wrote PVS-Studio technical author Egor Pishii in a post. The switchover will also help unify telemetry data collection a bit more. Preparing Java 28 The remaining production-ready feature is in-process data redaction (JEP 536), in which Java’s Flight Recorder removes all sensitive details before the recording is finalized. The previews that come with this release include lazy constants (JEP 531), pattern primitive types (JEP 532), structured concurrency (JEP 533), and a set of cryptographic objects (JEP 538). There’s also a vector API (JEP 537) still in the incubation stage. We’ll see all of these again, hopefully in a more polished form, in a later release. Work is already underway for Java 28, due in March 2027: Expect a simple JSON API as an incubating project as well as a revised Shenandoah garbage collector. ®

TypeSafe AI, a startup bestowed with $40 million in funding, on Tuesday declared itself a maker of frontier models with the release of Jev, a new type of AI model intended for interaction with machines rather than people. Instead of returning natural language in response to a text-based prompt, TypeSafe AI designed Jev to return typed probabilistic decisions for use by other software or AI models. Type safety in programming is a way to catch errors that arise when software processes an unexpected type of data, as might happen if an operation tried to divide an integer by a text string. By providing typed, structured values, Jev can avoid the parsing and validating that must be done to process text responses from LLMs. This approach can be useful for scenarios where interactions with an AI system need to be constrained to a limited set of answers. It can, for example, play Doom, when fed structured data describing the player's game state. But it's more likely to be used for sorting customer service problems and other business workflows. With Jev, the developer starts with a state value, which can be a JSON object or simply a string like "My card was charged twice." The model gets presented with state information through a set of question primitives (Choice, Score, Noul), each of which returns a different type of structured response with probabilities. So a question asking the model to choose which of three departments should handle a particular customer service query might come back: {"billing": 0.08, "technical": 0.85, "sales": 0.07}, with a confidence score of 0.82. That's not very useful for a person seeking the answer to a question but for a software developer, that might be just the thing to make a particular customer service function run more reliably. Jev is a System One model, which relies on a different architecture called Reinforcement Learning for Calibrated Decisions (RLCD). Diogo Almeida, co-founder and CEO of TypeSafe AI, is a former OpenAI researcher and one of the co-inventors of reinforcement learning for human feedback (RLHF) and ChatGPT. "TypeSafe was founded to pursue an alternative path for AI research, focused on machine-native AI," said Almeida in a statement. "I spent years working on models designed to make AI better at interacting with people. But if AI is going to fundamentally change how work gets done, people can't be the only consumers of intelligence." While natural language responses from the likes of ChatGPT or Claude can be wrapped in other code to produce a constrained set of responses, Jev's appeal follows from its efficiency. TypeSafe claims that LLMs generally have variable response times that can stretch into minutes. Jev's response time supposedly ranges from 70ms-500ms, 40x-200x faster than traditional LLMs. The model achieves this speed by using parallel processing. LLMs like OpenAI's GPT-5.6 Terra predict the next text token in a sequence and thus operate sequentially. The System One architecture returns all outputs to a query at once. The demo posted on the TypeSafe website shows Jev returning a response in 0.114s, compared to 8.566 seconds for OpenAI's GPT-5.6 Terra. Then there's the issue of cost. Jev charges $0.042 / MTok for input and $0 for output. OpenAI GPT-5.6 Terra costs $2.00 / MTok and $12 / MTok for output. And compared to a top tier model like Fable 5.1, Jev costs 238x less. The name "Jev" is a reference to 19th century English economist William Stanley Jevons, known for Jevons paradox, who argued that technology efficiency related to coal usage increased coal consumption rather than reducing it. The name also reflects the AI industry's bet that greater token efficiency will increase token consumption even as token prices decline. That wager assumes the market for tokens will be as broad as the market for energy – an assumption that isn't yet settled given that many people who have access to AI tools just don't have a use for them or avoid AI tools due to moral objections. TypeSafe claims that Jev is hallucination-free, which really isn't a fair comparison as its output is not natural language. Jev instead returns structured responses with probabilities, and that does not preclude the possibility of being incorrect. TypeSafe's model won't be the source of misinformation like fabricated legal citations but it also won't be used in the same way since its responses are intended to be consumed by software. Where this becomes relevant is when handling AI agent tool calls and automation. "Having a hallucinated tool call is inconvenient in an agent, but is an absolute deal-breaker if it’s part of a system with latency guarantees or it’s buried several layers deep in a dependency chain," the company claims in its blog post. "Existing models, no matter how smart, still hallucinate and have type errors." Jev, TypeSafe claims, is well-suited for AI automation software, real-time applications that need quick decisions, AI map reduce jobs for classifying large corpora of data, verification of AI inputs, and AI model harnesses. And if it can play Doom as well and as efficiently as advertised, it may not be long before users ask Jev to make drone targeting decisions. ®

Apple has addressed more than 260 CVEs across all of its operating systems, browsers, and other software products, marking the largest single patch cycle in Cupertino's history. While this CVE count is hardly notable compared to some vendors - hello, Microsoft’s record-breaking 974 bugs disclosed earlier this month - it does set a company record for Apple. It also reflects the new reality of AI-driven bug hunting, as models become exponentially better and faster at finding security vulnerabilities. However, the flip side of the AI coin we were promised - that models would also excel at writing patches and automatically fixing software and systems - yeah, that hasn't happened yet. The silver lining for everyone updating their Apple products right now (including this humble vulture): none of the vulnerabilities are listed as being under active exploitation. Of course, that may change very quickly as attackers are, at this very moment, looking to exploit the newly disclosed bugs, too. And we promise you that they are using AI. Apple’s latest mobile and operating system versions, iOS 27 and macOS 27 Golden Gate, released on Monday, also address a record 122 and 204 security vulnerabilities, respectively, across phone, iPad, and computer operating systems. Of these hundreds of CVEs, however, there are only ten (by our count) that AI is directly credited with finding. iOS 27 fixes 122 flaws Just two of the iPhone and iPad CVEs fixed with iOS 27 credit a coding agent or AI assistant with finding them. These include CVE-2026-65410, a vuln that exists in iPhone and iPad AVE video encoders, that can cause unexpected system termination. Apple credited AI-bug-finding firm Calif, along with Claude and Anthropic Research, with finding and reporting this security flaw. Then there's CVE-2026-65409, a type-confusion issue in iOS’ Foundation framework that can be abused to cause a denial of service, also found by Calif - specifically human researcher Bruce Dang - in collaboration with Claude and Anthropic Research. Some of the more interesting and serious iOS bugs fixed with the newest update aren’t listed as found by AI. These include CVE-2026-43689, a privilege-escalation flaw that could allow an app to gain root access. Apple credited Nosebeard Labs’ Andreas Jaegersberger and Ro Achterberg with reporting this bug. Additionally, CVE-2026-65406, a logic-issue flaw due to improper validation in Background Assets, could be abused to access sensitive user data. Background Assets is an Apple framework that lets you download large files and content in the background before a user opens the app for the first time. Baidu Security researcher Ye Zhang spotted this one. macOS 27 patches 204 vulns Meanwhile, the new macOS 27 update that addresses 204 vulns also fixes both the AI hunted bugs: CVE-2026-65410 and CVE-2026-65409. Plus, it credits AI helpers with discovering eight others, including one especially nasty flaw that could lead to remote code execution through the CUPS printer interface. Let’s start with that one. CVE-2026-43692 is a validation issue in CUPS that can be exploited by a remote user to either terminate the app or execute malicious code. Aaron Grattafiori and the Nvidia AI Red Team receive credit for disclosing this flaw. CVE-2026-64790 in CUPS can be exploited to gain elevated privileges. Grattafiori and the Nvidia AI Red Team again get credit for the win. CVE-2026-43791, a validation issue in StorageKit, can be abused to read files. Grattafiori, the Nvidia AI Red Team, Meridian Miftari, and Amy from amys.website disclosed this flaw to Apple. CVE-2026-43690 is a race-condition bug in the Server Message Block (SMB) network communication protocol. A local user can exploit the flaw to read kernel memory. Calif’s Bruce Dang, with Claude and Anthropic Research, found this one. CVE-2026-43719 is another SMB use-after-free bug, discovered by Calif’s Dang and Jakob Pammer, Claude, and Anthropic. “Mounting a maliciously crafted SMB network share may lead to system termination,” Apple warned. CVE-2026-65376 is yet another SMB issue - this one an out-of-bounds-read flaw reported by Dang, Claude, Anthropic, and 재영 정. CVE-2026-65374 is a memory-corruption issue in the WebDAV protocol that can lead to code execution. Dang, Claude, Anthropic, and He Wei (ギカク) receive credit for finding this bug. CVE-2026-65375, also in WebDAV, can cause unexpected system termination. Apple credited Dang, Claude, Anthropic, and Devcore Research Team’s YingMuo. CVE-2026-43677 is an out-of-bounds write issue in WebDAV with a slew of researchers receiving credit for finding it. In addition to the usual trio (Dang, Claude, and Anthropic), bubu, Omar Cerrito, HE WEI（ギカク), Roman Zabicki, Richard Zana, Chris Bailey - Short Circuit, Aswin Kumar Gokulakannan, and Surya Narayan Kushwaha are on the list.®

As Nvidia expands its influence through its NVLink Fusion tech, rival networking vendors are scrambling to bring alternative interconnects and switches to market. At the AI Infra Summit this week, Delos Data and Cornelis Networks officially entered the scale up networking race. Scale-up fabrics, like NVLink, are what have allowed Nvidia to make eight, 72, and now 576 GPUs behave as one enormous AI accelerator. To catch up, rivals like AMD have embraced emerging protocols like Ultra Accelerator Link. Today, these protocols are largely being tunneled over standard Ethernet switches. For instance, AMD is using Broadcom’s 102.4 Tbps Tomahawk 6-based switches connecting to custom I/O dies on the MI455X. Purpose-built UALink switches and physical interconnects remain elusive, but that won’t be the case for long if Cornelis and Delos have their way. The two companies are approaching this challenge from a few different angles, including standardization, software optimization, and physical hardware. Setting the standard for the Never-Nvidia network At AI Infra on Monday, HPC-centric networking vendor Cornelis introduced the Active Compute Fabric (ACF), which seeks to establish an open architecture for scale up and scale out networking that integrates programmable compute into the fabric. The standard signals Cornelis’ entry into the scale up networking arena. Spun out of Intel in 2020, Cornelis’ Omni-Path tech was originally designed as a scale-out interconnect for high-performance computing applications, including supercomputers like Trinity and Lynx. With the imminent launch of the company’s 800 Gbps-capable CN6000-series switches and NICs, the company has its sights set not only on bringing its tech to a broader audience through the open Ultra Ethernet protocol, but also on scale-up networking. ACF expands on the mission of technologies like UALink and the Ethernet for Scale-Up Networking, another scale up networking protocol, to set a baseline for in-network compute capabilities across a wide range of hardware, not just Cornelis’ own CN-series parts. One of the key technologies behind Nvidia’s NVSwitch ASICs is support for SHARP, which allows for things like in-network collectives to be offloaded to the switch ASICs, freeing up GPU compute and cutting down on communication overheads. Collective acceleration isn't new by any means, with major networking vendors from Broadcom to Cisco having implemented it in some capacity. As an open architecture, ACF appears to set a least common denominator so that customers deploying these systems know exactly what they’re getting. However, it doesn’t stop there. Cornelis sees opportunities to accelerate a variety of other functions. We’ve detailed a few below: KV cache offloading used to store and retrieve model state across multiple sessions. MoE expert dispatch to reduce duplicate transfers and communication overhead for mixture-of-experts models during inference and training. Message passing interface offload for HPC-centric applications. In fabric checkpointing for failure recovery during training and other large workloads. “Communications overhead and synchronization can leave expensive accelerators underutilized,” the company explained. These in-network accelerators “allow the network to operate on data as it moves through the system.” According to Cornelis, the savings potential from reclaiming that idle compute is significant. “In a 100,000 GPU system, Cornelis modeling of public data shows that roughly half of all GPU hours are spent waiting for data, worth about $1.68 billion a year in wasted capacity and 500 GWh of power.” As usual, take these claims with a grain of salt. However, reclaiming GPU idle time would equate to significant savings. Eliminating all the bottlenecks that contribute to them is easier said than done. A new kind of NIC for the AI age While Cornelis champions its ACF architecture, Delos Data, a startup founded by former Barefoot Networks and Intel execs, aims to tackle the physical layer with a series of new network reference designs. Marketed under its Nonstop AI portfolio, the network interfaces span the full gamut of connectivity from co-packaged interconnects to more traditional NICs. The idea, the company explains, is to provide customers with a system-level blueprint for speeding up data movement between endpoints and enabling larger compute domains scaling beyond a single rack. Delos’ data interface will be offered in three form factors. The first is an I/O die capable of more than 30 Tbps of aggregate bandwidth or about 8 TB/s in either direction. That’s more than double the interconnect bandwidth of either Nvidia's or AMD’s latest accelerators, which cap out at 3.6 TB/s. Just how competitive that ends up being will depend on when Delos’ I/O chiplets actually see deployment. That timeline is going to depend heavily on integration since those I/O dies need to be integrated directly into the accelerator package, which requires a high-degree of co-design. Critically, Delos isn't trying to trap its customers in a walled garden to the same extent that Nvidia does with its NVLink Fusion I/O dies or IP. At least as of writing, NVLink Fusion still requires customers to buy NVSwitches for scale up networking. Delos’ chiplets are protocol agnostic. They don’t care whether you’re using ESUN, UALink, or something else. Chip designers that'd rather focus their efforts and capital on the AI bits of their accelerators – which, it turns out, applies to most hyperscalers – could simply license Delos' chiplet design. This protocol agnosticism also extends to Delos’ near-packaged optics (NPO) tech, which will integrate a 10-plus Tbps data interface — that’s 2.5 TB/s bidirectional bandwidth in Nvidia speak — with optics engines from leading optics suppliers. Rack scale systems, like Nvidia’s NVL72, have largely relied on copper interconnects up to this point due to power constraints. As systems grow from 72 GPU rack systems to row-scale clusters with 576 and eventually 1,152 GPUs, optics become unavoidable. While integrating optics directly into the accelerators is possible using tech available today, the blast radius of a failed optical module is considerable. NPO offers an alternative. Copper interconnects are used within the rack, with optics provided via user-serviceable NPO modules between racks. Finally, Delos is working on a 400-plus Gbps NIC, which like the rest of its data interface offerings is protocol agnostic. This is arguably the least surprising entry in the entire lineup. Scale out networks are still going to be required for large training clusters as well as for front-end access and storage networks. These interfaces build on Delos’ existing compute reference design, which we looked at during Computex, as well as its Nonstop AI software platform, which is designed to facilitate the configuration and monitoring of these switched fabrics or meshes in order to enable dynamic rerouting of traffic in the event of a link failure. Telemetry gathered by its data interface offerings allows for even greater visibility into the network, allowing for faster rerouting and recovery. Fueling competition Investors appear more than eager for more competition in the scale up arena. Alongside its product announcements, Cornelis announced approximately $205 million in funding to bring its next generation of scale up and scale out networking products to market. Meanwhile, Delos Data announced Tuesday that it’s now raised more than $100 million in funding with support from venture capital firms Matrix, Playground, and Socratic Partners among others. ®

AWS engineers have published an open-source AI agent management tool that treats tasks and permission requests just like emails that fill up an inbox. Pizza Bot was announced on AWS’ open-source blog last week, where it was billed by its developers as a way to more easily manage agent tasks without constantly having to check a chat window to see if a task was completed or on hold because more permission was needed. “You don’t send an email and then sit watching the outbox until the reply lands,” the Pizza Bot team wrote. “Pizza Bot is shaped like an email client for the same reason: a thread is a unit of work you come back to rather than a session you have to attend.” Like the modern email services it mimics, Pizza Bot organizes communications with AI agents into message threads. Finished tasks, along with detailed result writeups you’d expect from a colleague you’ve delegated your work to, end up in the Unread category, while anything you need to make a decision on is shunted to the Action category. Asynchronous agent communication is likely to be a godsend for those balancing multiple tasks assigned to digital delegees, as the Pizza Bot team describes it. “Live chat assumes both parties are present, which holds for a quick exchange and breaks the moment a task takes several minutes,” the Pizza Bot maintainers wrote. Just like you don’t want to stare over the shoulder of the poor IT sod you sent off to do a task while you watch YouTube videos, no one wants to sit there and watch an AI agent do web research, or whatever. “Pizza Bot treats an agent the way you’d treat a colleague who has gone off to do the work: it comes back when there’s something to read, or something only you can decide,” the PB team added. Internal tool goes open source Pizza Bot, named for Amazon’s “two-pizza team” model (i.e., small, autonomous teams small enough to be fed by two pizzas), began life inside Amazon as an internal tool for AI agent uses other than software development. According to its developers, non-coding uses they were exploring internally continually had the same requirements, namely working in the background autonomously and only bugging their user when they were done or had a question. “None of those describe how anyone uses a chat window,” the Pizza Bot team explained. “They describe how people already work with each other.” Email, they added, was the perfect design paradigm for internal use, and seeing whether it would work for everyone else meant opening the project to open source contributions. As mentioned, Pizza Bot is open source - you can snag a copy for Windows, macOS, and Linux from its GitHub repo. It runs on-device, stores all its communication threads locally in SQLite files, and doesn’t transmit anything off device unless you grant permission for it to do so. Prompts and attachments are sent to model providers, naturally, but even that’s not necessary: Pizza Bot also supports Ollama, allowing users to run compatible open-weight AI models locally. Pizza Bot doesn’t ship with a lot of capabilities, as its maintainers noted that Amazon-specific skills and MCP servers were left out when they rebuilt it as an open source project for public release. Out of the box, however, it can list, read, write, edit, and search files, delegate tasks to specialist agents, and run a sandboxed JavaScript interpreter. It also ships with a browser automation skill and a guide skill that lets users ask about Pizza Bot’s capabilities. Claude Code-compatible .mcp.json files are supported and can be dropped straight in, as can skills, the team noted. One word of caution to anyone who wants to try it: The devs make clear this is a piece of open source software, not an officially supported AWS product. “There’s no AWS support or service-level agreement behind it,” they note. If you’re gonna bake a pie with this bot, you’ve been warned. ®

If your employees are visiting Chinese-language gambling or adult sites, they may not just be wasting time and money, but potentially encountering serious malware hidden behind domains that look like mostly harmless entertainment at first glance. A report from Infoblox urges the security community to pay closer attention to these websites, because some double as command-and-control (C2) infrastructure for espionage and malware distribution. Zach Edwards, staff threat researcher at Infoblox, suggests security researchers and the media have ignored these sites because the story is complicated and confusing. Infoblox says it tracks about 1.7 million Chinese-language casino websites that facilitate illegal gambling. These support North Korean money laundering and tax avoidance, among other dubious activities. And these casino sites can be difficult to distinguish from one another. They tend to use variations of common templates in terms of design and function. Many operate like a legal casino would, just relying on the advantage of house odds to profit. While these sites provide illegal gambling and adult entertainment for online visitors from China and Asia, some rely on US cloud providers for computing infrastructure. "Major US hosting companies (Amazon, Microsoft, Cloudflare, and Google) continue to host infrastructure associated with these domains," the Infoblox report explains. "One likely explanation is account theft at those providers, a practice documented previously as 'infrastructure laundering.'" That refers to hosting companies like Funnull that have reportedly rented IP addresses from Amazon Web Services and Microsoft and made those resources available to clients carrying out illegal activities. According to a July 2026 report from the UN Office on Drugs and Crime (UNODC), disparate crime syndicates increasingly use common infrastructure for cybercrime, while online scams resulted in estimated losses of between $88.3 billion and $114.1 billion in 2025 across East Asia, Southeast Asia, Australia, and New Zealand. A subset of casino sites offer scam gambling, or "scambling." Visitors place bets but can't get their money out if they win. And then there's a subset of sites used by China-aligned threat groups. "China-aligned APT groups have been running the PeckBirdy framework since 2023, hiding their malware C2 domains inside low-quality Chinese-language casino websites," Infoblox said. PeckBirdy, as noted by Trend Micro researchers in January, is a script-based framework that attackers can load through compromised websites. In one campaign, attackers injected scripts into gambling sites that loaded PeckBirdy and displayed fake software update pages designed to entice victims to download malware. The problem is that each of these three types of sites, though they change frequently, looks similar. Infoblox notes that just over 3 percent of its enterprise customers resolved at least one PeckBirdy C2 domain. "The most important thing for defenders to do is stop ignoring casino domains," Infoblox argues. "An alert on a Chinese-language casino or adult domain that gets closed as an employee browsing violation is precisely the outcome the PeckBirdy operators are counting on. The decoy works because the dismissal is reasonable – these domains genuinely are, most of the time, exactly what they appear to be." Security analysts who review suspicious network contacts are advised to check whether these casino domains include malicious payloads before closing the review ticket. ®

Iranian state cyber actors are targeting individuals using social messaging apps to deploy surveillance and data-stealing malware on their Windows machines, three Western governments warned. In all observed cases, Chosen Brick has infected Windows systems exclusively. Iran has used it since at least 2025 to take over individuals’ devices, stealing their contacts, emails, and social media messages, which allows the spies to track people’s movements, the FBI, UK National Cyber Security Centre, and the Netherlands’ General Intelligence and Security Service (AIVD) said on Tuesday. “Iran almost certainly uses cyber activity to support the repression of individuals who are seen as a threat to the regime, such as dissidents, activists and journalists,” the security advisory said. “In some cases, the Iranian intelligence services have plotted to kidnap or conduct lethal operations against individuals internationally, who they perceive as enemies of the regime.” These attacks typically begin with WhatsApp and Telegram messages, purportedly coming from individuals and organizations that the victim knows and trusts. The Iranian spies do a significant amount of research to prepare for these social engineering campaigns. By the time they send the initial message via a social media app, they have “extensive” knowledge of the targeted individual, their contacts, and relevant industry organizations to make the phony messages more believable, according to the agencies. After building rapport with the mark, the attackers convince them to download and open a file that appears to be a legitimate application. Specifically: Pictory, RunwayML, Norton Antivirus, Telegram, Adobe Flash Player, and KeePass are among the legitimate applications the malicious files have been made to resemble, the government agencies said. Upon opening the file, the malware executes without the victim’s knowledge, and will survive a reboot of the target device. Chosen Brick also adds exclusions to Microsoft Defender antivirus in an attempt to evade detection, and then connects to Telegram for command-and-control (C2) communications using a victim-specific Telegram bot. While the malware hasn’t yet been observed to automate lateral movement across the network, this is “technically possible,” the advisory noted. It does, however, download additional malware and set up persistence for new payloads on infected devices, using the same registry key that Chosen Brick uses to establish its own persistence on a Windows device: HKCU\Software\Microsoft\Windows\CurrentVersion\Run. Other features include enumerating running processes and system information, capturing screen and audio content, stealing emails, along with Telegram and WhatsApp data from web browsers, and wiping the computer system. “Organizations that are concerned Chosen Brick has been executed should contact their IT providers, either internal or external, to investigate,” the US, UK, and the Netherlands warned. “As this actor targets personal devices, not just corporate devices, organizations are recommended to circulate this with their staff that are likely to be targeted and support them in checking their personal devices too.” The Western agencies’ latest Iran alert follows a series of water and energy cyberattacks that researchers and media reports have linked to Iran, although the US and UK governments have stopped short of formally attributing them, as the military conflict between Iran and the US approaches its seventh month. In August, America’s lead cybersecurity agency, CISA, disclosed that the July cyberattacks that disrupted American water utilities across 12 states targeted more than 100 internet-exposed water systems. CISA did not, however, attribute the campaign to Iran or anyone else. Around the same time, a suspected Iran-linked cyberattack also shut down a small UK power plant. Also in August, five US agencies warned that attackers are using AI-generated exploitation scripts to break into internet-exposed Siemens S7 Series programmable logic controllers (PLCs) at water, manufacturing, energy, and other critical facilities. “This is not a theoretical risk – it is an active threat,” the feds warned. ®

The US Department of Energy has selected a new participant for a program aimed at expanding the domestic supply of fuel for next-generation nuclear reactors. However, we shouldn't expect that to mean the US' ever-growing fleet of datacenters will be powered by low-carbon nuclear energy any time soon. Utah-based Nusano announced on Monday that the DoE had selected it for its Nuclear Energy Launch Pad Program designed to help companies commercialize their nuclear power concepts. In this case, Nusano is working on a new method of creating high-assay low-enriched uranium, or HALEU, fuel that’s required for most next-generation nuclear reactors, like small modular reactors (SMRs) and molten salt reactors. HALEU is defined as uranium enay, low-riched to between 5 and 20 weight percent of U-235, the primary fissile isotope that nuclear reactors use to generate power. Traditional nuclear reactors use low-enriched uranium of under five percent. The higher enrichment is needed for smaller, more efficient reactor designs that are currently under development. Several advanced reactor demonstrations in the US have reached zero-power criticality, but none is yet operating commercially. The US has been without a steady commercial-scale supply of HALEU for years, with Russia historically the main commercial supplier and China also possessing HALEU enrichment capability. As of 2026, there’s only a single company in the US actively producing HALEU, and it’s not Nusano. The DoE has signed agreements with several others to produce more HALEU and related fuel products, but those firms are still establishing their facilities, and production remains minuscule. Centrus, the only American company producing HALEU, has generated less than two metric tons since it signed a contract with the government in 2019. The competition heats up Nusano expects to blow Centrus’ meager fuel generation out of the water with its HALEU enrichment design. That said, the company’s design hasn’t actually been realized yet. Nusano didn’t answer a question about whether it had built a functional demonstration HALEU enrichment unit, and when asked whether it had produced any HALEU, the company said that prototype development is still under way. “Nusano expects each unit will produce 5.9 metric tons (MT) of HALEU per year, with the first unit scheduled to be operational in 2031,” Nusano comms boss Scott Larrivee told The Register in an email. “Nusano’s Launch Pad application includes a well-developed, practical plan to move from proof-of-concept to commercial implementation.” As for what that’ll look like, Nusano describes a method that’s completely different from other forms of HALEU production. Whereas other HALEU refiners use gas centrifuges to produce HALEU, spinning uranium hexafluoride gas at high speeds to increase U-235 concentration, Nusano’s method ditches gas entirely. “The method is complex, costly, and potentially hazardous,” Larrivee explained. “If released, uranium hexafluoride reacts with moisture in air or human tissue to form highly toxic, corrosive hydrogen fluoride and uranyl fluoride.” Those gases, Larrivee said, can cause potentially fatal injuries like chemical burns and lung and kidney damage. Nusano is proposing a method that uses uranium feedstock that’s converted to metal prior to enrichment. Unspecified “proprietary separation techniques” are then used to enrich the metal and separate out U-235. The HALEU output of Nusano’s design is metal. “The Nusano process is metal in, metal out. No deconversion. No gaseous fluorine chemistry anywhere,” Larrivee added. Nusano’s current design only requires 1,200 square feet per HALEU unit, and the company anticipates its production cost will be lower than those of gas-centrifuge enrichment. That said, a lot of Nusano units will be needed to meet expected American demand. According to the DoE, HALEU demand could reach 50 metric tons per year by 2035, and Nusano is only expecting to get its first 5.9 MT/year unit up and running by 2031. It’s also not clear if Nusano’s current promises will still be the same if and when its HALEU units are up and running. In June 2025, the company said it had achieved a breakthrough that would allow a single HALEU unit to produce 50 metric tons annually and enable it to scale production to 350 metric tons per year by 2029. Search the company website for that announcement today, and it’s gone. We were only able to see it by pulling up an archived copy. Larrivee said that announcement was out of date, developed under prior company leadership, and that the numbers in the latest release were reflective of current development plans. A number of datacenter operators have promised to embrace SMRs and other next-gen nuclear when the technology is available. With HALEU output still low and commercial-scale production years away, it's unclear when those promises will be realized. Meanwhile, the Trump administration's Environmental Protection Agency has just scrapped Biden-era emissions regulations for coal and gas power plants, which it said means coal generation in the US could increase by more than ten times current levels. In other words, next-gen nuclear is, as ever, still on the horizon while bit barns crank up the fossil fuel turbines, now with even less regard to environmental impacts. ®

Financial service biz Moody's says the availability of power is still a major constraint on US datacenter growth as the pace of construction exceeds the speed at which the nation's grid infrastructure can add the requisite capacity. America is the epicenter of the datacenter construction boom, driven by the AI craze and demand for compute capacity to train ever-larger foundation models. The latest report from Moody's estimates the power needed by all those bit barns will reach 426 terawatt-hours (TWh) in 2030, citing data from the International Energy Agency (IEA). That is almost double the 2025 figure. It falls on the power sector to develop new resources to meet this extra demand as traditional power generation is struggling to keep pace. Moody's calculates that building the additional generation capacity will cost about $110 billion, adding an estimated $25 billion to $30 billion a year to electricity system costs. Yet efforts to supply the rapidly growing number of cloud and AI server farms are running up against lengthy permitting processes and long lead times for key equipment and materials, causing delays of up to seven years in some cases. Management consultancy Bain & Company highlighted the problem nearly two years ago when it warned that energy companies needed to significantly boost their generation capacity, reversing years of flat or declining demand. Transmission buildout is a parallel bottleneck, with $80 billion to $115 billion in grid investment through 2030 already approved or under construction across US power markets, Moody's says, with the potential to triple over the next decade. Investment bank Jefferies reported in June that half of the extra US datacenter capacity planned for 2026 is unlikely to come online this year, blaming power availability and grid connection setbacks, along with other issues including zoning and permitting challenges. Regions able to accelerate development, such as Texas, have experienced a wave of datacenter construction. The resulting pressure on local grids has prompted some authorities to clamp down on new connections for compute sheds. Moody's believes most new IT campus operators are still likely to seek grid connections to limit development costs and benefit from service reliability and shared redundancy. Yet many are also seeking alternatives, including on-site behind-the-meter (BTM) power generation to avoid delays in bringing facilities online. The document also notes that the growing energy demands from server farms are inflating power prices and customer bills in some regions of America, as The Register reported earlier this year. This has led to affordability concerns, fueling opposition to datacenters and prompting measures such as stricter approval frameworks and moratoriums on new projects. President Trump even introduced a Ratepayer Protection Pledge, under which tech giants and energy companies committed to shielding consumers from rising energy costs due to the datacenter boom. In the long term, Moody's believes that large-scale infrastructure investment will ultimately strengthen grid reliability and affordability, despite the risks. Extra generation capacity will alleviate longstanding supply constraints, it says, reducing price volatility during periods of extreme weather or high demand, while expanded transmission networks could improve access to lower-cost generation resources, particularly when datacenter developers or their tenants help finance the associated infrastructure. ®

Criminals are exploiting a critical Cisco Secure Email Gateway flaw that can turn a malicious email into root access. The vulnerability, tracked as CVE-2026-76461, carries a 9.8 CVSS score and affects physical and virtual Secure Email Gateway appliances regardless of their configuration. Cisco says there are no workarounds, so patching is the only fix. The bug lies in how Cisco's AsyncOS software handles incoming email. An attacker doesn't need to log in: they can send a booby-trapped message through a vulnerable gateway and, if the exploit works, run commands as root. Which is not exactly what you want from the box tasked with keeping nasty emails out. Cisco's Product Security Incident Response Team said it became aware of active exploitation in September, although the networking giant hasn't said who is behind the attacks, how long they have been going on, or how many organizations have been compromised. Cisco uncovered the bug while resolving a Technical Assistance Center support case. Signs suggest at least some Cisco cloud customers were caught up in the attacks. Cisco said it investigated devices belonging to its Secure Email Cloud service and directly contacted customers whose appliances showed indicators of possible compromise. It is now carrying out remediation and recovery work and says all of its Secure Email Cloud devices have been upgraded to AsyncOS 16.5.0-780. Admins running their own appliances have a little more work to do. Cisco recommends checking logs for signs of suspicious activity, but warns that finding nothing doesn't necessarily mean the system is clean. Once attackers have root access, Cisco says they could tamper with the logs and cover their tracks. Admins are also being told to check network and firewall logs for anything unusual, rather than relying on the gateway itself for answers. For virtual appliances suspected of being compromised, Cisco's recovery advice is fairly drastic: preserve the forensic evidence, deploy a fresh VM running fixed software, rebuild the configuration, and rotate credentials and cryptographic material. Cisco has fixed the bug in AsyncOS releases 15.5.5-014, 16.0.4-302 and 16.5.0-780, with customers strongly encouraged to move to the latter. There's still a decent-sized target pool out there. The Shadowserver Foundation was tracking more than 400 Cisco Secure Email Gateway appliances exposed to the internet as of Monday The flaw has also landed in CISA's Known Exploited Vulnerabilities catalog, with US federal civilian agencies ordered to remediate it by September 17. CVE-2026-76461 comes less than a year after attackers exploited another critical AsyncOS flaw, CVE-2025-20393, to break into Cisco Secure Email Gateway appliances and install persistence mechanisms. That bug eventually scored a perfect 10. For anyone still running an affected gateway, the message is fairly simple: the box designed to inspect hostile email can itself be pwned by one; attackers are already doing it, and there is no workaround to hide behind. ®