joshuark shares a report from Reuters: Microsoft has filed an amicus brief on Tuesday in support of Anthropic's lawsuit asking the court to temporarily block the U.S. Department of Defense designation of the AI startup as a supply-chain risk. In an amicus brief filing in a federal court in San Francisco, Microsoft backed Anthropic's request for a temporary restraining order against the Pentagon order, arguing that its determination should be paused while the court considers the case. Microsoft, which integrates the AI lab's products and services into technology it provides to the U.S. military, said that it was directly impacted by the DOD designation.

"Should this action proceed without the entry of a temporary restraining order, Microsoft and other government contractors with expertise in developing solutions to support U.S. government missions will be forced to account for a new risk in their business planning," the company said. Microsoft's filing argued the TRO is needed to prevent costly disruptions for suppliers, who would otherwise have to rapidly rebuild offerings that rely on Anthropic's products. The judge overseeing the case must approve Microsoft's request to file the brief before it is officially entered, but courts often permit outside parties to weigh in on important cases.

BrianFagioli writes: Google says it will finally release Chrome for ARM64 Linux in the second quarter of 2026, bringing the company's full browser to a platform that has existed for years without official support. Until now, Linux users running Arm hardware have largely relied on Chromium builds or unofficial packages if they wanted something close to Chrome. Google says the new build will include the same features found on other platforms, including Google account syncing, Chrome Web Store extensions, built-in translation, Safe Browsing protections, and Google Password Manager.

The timing reflects how ARM hardware is becoming more common across the Linux ecosystem, from developer laptops to AI systems. Google also pointed to NVIDIA's DGX Spark, a compact AI supercomputing device built on the Grace Blackwell architecture, which will support installing Chrome through NVIDIA's package management tools. For many Linux users, the announcement feels like a "finally" moment, as ARM64 Linux systems have been widespread for years despite the absence of an official Chrome build.

Shantanu Narayen announced he will step down as CEO of Adobe once a successor is appointed, ending an 18-year tenure during which he transformed the company from boxed software to the Creative Cloud subscription model. Narayen said he will remain board chair as Adobe continues pushing into generative AI products. CNBC reports: Narayen joined Adobe in 1988 as a vice president and general manager, and he became CEO in 2007. Under Narayen, Adobe pushed from software licenses to subscriptions to its Creative Cloud application bundle, and the company is now working to expand through generative artificial intelligence. He sought to acquire fast-growing design software company Figma, but regulators pushed back, and the companies called off the deal, resulting in Adobe paying Figma a $1 billion breakup fee. [...]

Narayen, 62, is lead independent director of Pfizer in addition to his responsibilities at Adobe, where he received $51 million in total compensation for the 2025 fiscal year, according to a filing. He owns $118 million in Adobe shares, according to FactSet. [...] On Narayen's watch, Adobe's stock jumped more than sixfold, while the S&P 500 is up about 350% over that stretch. "What attracted me to Adobe 28 years ago was our leadership in creating new market categories, world-class products, a relentless desire to innovate in every functional area of the company and the people I met during the interview process," Narayen wrote. "We have continued to create new markets, deliver world-class products, drive innovation in everything we do and attract and retain the best and brightest employees."

Perplexity AI has introduced a "Personal Computer" agent system that can run on a local machine such as a Mac mini, giving its AI agents access to a user's files and applications to automate tasks. According to CEO Aravind Srinivas, the heavy AI processing runs on Perplexity's "secure servers" but sensitive actions will require user approval. There will also be activity logs and a kill switch available to help ease concerns. AppleInsider reports: Perplexity Computer is, effectively, an AI that is a go-between for other AIs. Instead of issuing specific instructions to multiple AIs, you provide the general outcome of the task to Perplexity Computer. Perplexity Computer then breaks down the task into subtasks, which it then provides to sub-agents to do the actual work. In effect, you're talking to a project manager, who then delegates the task to other AIs, before combining the results and presenting them to you.

The managing AI has a lot more freedom in how it orders its subordinates than users may think. While one may create documents while another gathers data, the manager may go as far as to order the creation of software to complete its tasks. Personal Computer is an extension of this, in that it is a locally run app that ideally runs on a Mac mini. The app gives always-on, local access to the Mac's files and apps, which Perplexity Computer and the Comet Assistant can use and alter if required.

Anthropic updated Claude so it can automatically generate charts, diagrams, and other interactive visualizations directly inside conversations, rather than only in a side panel. The new visualizations are rolling out now to all users. The Verge reports: As an example, Anthropic says a conversation about the periodic table could lead Claude to generate a visualization of it, featuring interactive elements that let you click inside the table for more information. Another example shows how Claude can generate a visual related to a question about how weight travels through a building. Though Claude will automatically determine whether it should generate a visualization in your chat, Anthropic notes that you can also ask the chatbot to generate a diagram, table, or chart directly. [...]

Anthropic already allows you to create charts, documents, tools, and apps through Claude's "artifacts" feature, which opens in a side panel where you can interact, share, and download the AI-generated creation. But, as noted by Anthropic, artifacts are persistent, while the visualizations created within Claude's conversations will change or disappear as the conversation progresses. You can also ask Claude to make changes to the visualizations it creates.

Google Maps is rolling out its biggest update in more than a decade, introducing a Gemini-powered chatbot and a new "Immersive Navigation" interface. "Ask Maps" lets users plan trips, ask questions, and refine travel suggestions conversationally within the app. "The new chatbot will be accessible via a button up near the search bar," notes Ars Technica. "You can ask it anything you're likely to find in Google Maps without jumping into another app. You can ask for directions, of course, but it can also plan out road trips and vacations from a single prompt. Ask Maps works like a chatbot, so it accepts follow-up prompts to refine and expand on its suggestions."

Meanwhile, Google is promising a "complete transformation" of the navigation experience in Maps with what they're calling "Immersive Navigation." It brings detailed 3D visuals, smarter route previews, and improved guidance powered by data from Street View and aerial imagery. "You'll see accurate overpasses, crosswalks, landmarks, and signage in the new navigation experience," reports Ars. "Google also aims to solve some of the biggest usability issues with turn-by-turn navigation in this update. [...] Immersive Navigation tries to show you more of the route as you drive, using smart zoom and transparent buildings to help you plan ahead. Voice guidance will also reference turns after the next one where appropriate."

Immersive Navigation will also highlights the tradeoffs between different route options, such as longer routes that avoid traffic or tolls. And, as you approach your destination, it will uses Street View imagery, building entrances, and parking information to help you orient yourself. The features are launching on Android and iOS first, with broader platform support coming later.

An anonymous reader quotes a report from Bloomberg: Atlassian plans to cut 1,600 jobs or a 10th of its global workforce, joining rivals in slashing staffing to cope with the advent of AI and a broader post-Covid industry slowdown. Australian billionaire founder Mike Cannon-Brookes explained the reductions in a staff memo, while also announcing his chief technology officer was leaving the Sydney-based company. "It would be disingenuous to pretend AI doesn't change the mix of skills we need or the number of roles required in certain areas," Cannon-Brookes said. "It does."

Grammarly has disabled its Expert Review feature after backlash from writers whose names were used to present AI-generated feedback without their permission. Superhuman (formerly Grammarly) CEO Shishir Mehrotra wrote in a LinkedIn post that the company will disable Expert Review while they "reimagine" the feature: Back in August, we launched a Grammarly agent called Expert Review. The agent draws on publicly available information from third-party LLMs to surface writing suggestions inspired by the published work of influential voices.

Over the past week, we received valid critical feedback from experts who are concerned that the agent misrepresented their voices. This kind of scrutiny improves our products, and we take it seriously. As context, the agent was designed to help users discover influential perspectives and scholarship relevant to their work, while also providing meaningful ways for experts to build deeper relationships with their fans. We hear the feedback and recognize we fell short on this. I want to apologize and acknowledge that we'll rethink our approach going forward.

After careful consideration, we have decided to disable Expert Review while we reimagine the feature to make it more useful for users, while giving experts real control over how they want to be represented -- or not represented at all.

We deeply believe in our mission to solve the "last mile of AI" by bringing AI directly to where people work, and we see this as a significant opportunity for experts. For millions of users, Grammarly is a trusted writing sidekick -- ever-present in every application, ready to help. We're opening up this platform so anyone can build agents that work like Grammarly -- expanding from one sidekick to a whole team. Imagine your professor sharpening your essay, your sales leader reshaping a customer pitch, a thoughtful critic challenging your arguments, or a leading expert elevating your proposal. For experts, this is a chance to build that same ubiquitous bond with users, much like Grammarly has. But in this world, experts choose to participate, shape how their knowledge is represented, and control their business model. That future excites me, and I hope to build it with experts who want to develop it alongside us.

Nvidia is preparing to launch an open-source AI agent platform called NemoClaw, designed to compete with the likes of OpenClaw. According to Wired, the platform will allow enterprise software companies to dispatch AI agents to perform tasks for their own workforces. "Companies will be able to access the platform regardless of whether their products run on Nvidia's chips," the report adds. From the report: The move comes as Nvidia prepares for its annual developer conference in San Jose next week. Ahead of the conference, Nvidia has reached out to companies including Salesforce, Cisco, Google, Adobe, and CrowdStrike to forge partnerships for the agent platform. It's unclear whether these conversations have resulted in official partnerships. Since the platform is open source, it's likely that partners would get free, early access in exchange for contributing to the project, sources say. Nvidia plans to offer security and privacy tools as part of this new open-source agent platform. [...]

For Nvidia, NemoClaw appears to be part of an effort to court enterprise software companies by offering additional layers of security for AI agents. It's also another step in the company's embrace of open-source AI models, part of a broader strategy to maintain its dominance in AI infrastructure at a time when leading AI labs are building their own custom chips. Nvidia's software strategy until now has been heavily reliant on its CUDA platform, a famously proprietary system that locks developers into building software for Nvidia's GPUs and has created a crucial "moat" for the company.

YouTube is expanding its AI deepfake detection tools to a pilot group of politicians, government officials, and journalists, allowing them to identify and request removal of unauthorized AI-generated videos impersonating them. TechCrunch reports: The technology itself launched last year to roughly 4 million YouTube creators in the YouTube Partner Program, following earlier tests. Similar to YouTube's existing Content ID system, which detects copyright-protected material in users' uploaded videos, the likeness detection feature looks for simulated faces made with AI tools. These tools are sometimes used to try to spread misinformation and manipulate people's perception of reality, as they leverage the deepfaked personas of notable figures -- like politicians or other government officials -- to say and do things in these AI videos that they didn't in real life.

With the new pilot program, YouTube aims to balance users' free expression with the risks associated with AI technology that can generate a convincing likeness of a public figure. [...] [Leslie Miller, YouTube's vice president of Government Affairs and Public Policy] explained that not all of the detected matches would be removed when requested. Instead, YouTube would evaluate each request under its existing privacy policy guidelines to determine whether the content is parody or political critique, which are protected forms of free expression. The company noted it's advocating for these protections at a federal level, too, with its support for the NO FAKES Act in D.C., which would regulate the use of AI to create unauthorized recreations of an individual's voice and visual likeness.

To use the new tool, eligible pilot testers must first prove their identity by uploading a selfie and a government ID. They can then create a profile, view the matches that show up, and optionally request their removal. YouTube says it plans to eventually give people the ability to prevent uploads of violating content before they go live or, possibly, allow them to monetize those videos, similar to how its Content ID system works. The company would not confirm which politicians or officials would be among its initial testers, but said the goal is to make the technology broadly available over time.

An anonymous reader quotes a report from Bloomberg: Chinese authorities moved to restrict state-run enterprises and government agencies from running OpenClaw AI apps on office computers, acting swiftly to defuse potential security risks after companies and consumers across China began experimenting with the agentic AI phenomenon. Government agencies and state-owned enterprises, including the largest banks, have received notices in recent days warning them against installing OpenClaw software on office devices for security reasons [...]. Several of them were instructed to notify superiors if they had already installed related apps for security checks and possible removal, some of the people said.

Certain employees, including those at state-run banks and some government agencies, were banned from installing OpenClaw on office computers and also personal phones using the company's network, some of the people said. One person said the ban was also extended to the families of military personnel. Other notices stopped short of calling for an outright ban on OpenClaw software, saying only that prior approval is needed before use, the people said. The warning underscores Beijing's growing concern about OpenClaw, an agentic AI platform that requires unusually broad access to private data and can communicate externally, potentially exposing computers to external attack. [...]

Despite the potential security risks, companies from Tencent to JD.com Inc. have been rolling out OpenClaw apps to try and capitalize on the groundswell of enthusiasm, while several local government agencies have declared millions of yuan in subsidies for companies that develop atop the platform. [...] Tech giants like Tencent and Alibaba, along with AI upstarts ranging from Moonshot to MiniMax, have rolled out their own tweaks of the software touting simple, one-click adoption. A slew of government agencies, in cities from Shenzhen to Wuxi, have issued notices offering multimillion-yuan subsidies to startups leveraging OpenClaw to make advances. The frenzy has helped drive up shares of AI model developer MiniMax nearly 640% since its listing just two months ago. It's now worth about $49 billion, surpassing Baidu -- once viewed as the frontrunner in Chinese AI development -- in market value. The company launched MaxClaw, an agent built on OpenClaw, in late February.

ASUS says the MacBook Neo is a "shock" to the Windows PC ecosystem. "In the past, Apple's pricing situation has always been high, so for them to release a very budget-friendly product, this is obviously a shock to the entire industry," said ASUS co-CEO S.Y. Hsu in a Tuesday earnings call. While he expects PC makers to respond, rising AI-driven memory shortages could push hardware prices higher across the industry. PCMag reports: Hsu said he believes all the PC players -- including Microsoft, Intel, and AMD -- take the MacBook Neo threat seriously. "In fact, in the entire PC ecosystem, there have been a lot of discussions about how to compete with this product," he added, given that rumors about the MacBook Neo have been making the rounds for at least a year. Despite the competitive threat, Hsu argued that the MacBook Neo could have limited appeal. He pointed to the laptop's 8GB of "unified memory," or what amounts to its RAM, and how customers can't upgrade it.

He also described the MacBook Neo as a "content consumption" device, similar to an iPad. "This is different from the use case of a mainstream notebook," which can handle more compute-intensive tasks, Hsu said. "How big of an impact [the MacBook Neo] will have on the PC industry will still require some time for us to observe," Hsu said while suggesting it might not gain traction among Windows PC users due to software differences. "Of course, the entire Windows PC ecosystem will push out products to compete against Apple," he added.

An anonymous reader quotes a report from Wired: Advanced Machine Intelligence (AMI), a new Paris-based startup cofounded by Meta's former chief AI scientist Yann LeCun, announced Monday it has raised more than $1 billion to develop AI world models. LeCun argues that most human reasoning is grounded in the physical world, not language, and that AI world models are necessary to develop true human-level intelligence. "The idea that you're going to extend the capabilities of LLMs [large language models] to the point that they're going to have human-level intelligence is complete nonsense," he said in an interview with WIRED.

The financing, which values the startup at $3.5 billion, was co-led by investors such as Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions. Other notable backers include Mark Cuban, former Google CEO Eric Schmidt, and French billionaire and telecommunications executive Xavier Niel. AMI (pronounced like the French word for friend) aims to build "a new breed of AI systems that understand the world, have persistent memory, can reason and plan, and are controllable and safe," the company says in a press release. The startup says it will be global from day one, with offices in Paris, Montreal, Singapore, and New York, where LeCun will continue working as a New York University professor in addition to leading the startup. AMI will be the first commercial endeavor for LeCun since his departure from Meta in November 2025. [...]

LeCun says AMI aims to work with companies in manufacturing, biomedical, robotics, and other industries that have lots of data. For example, he says AMI could build a realistic world model of an aircraft engine and work with the manufacturer to help them optimize for efficiency, minimize emissions, or ensure reliability. LeCun says AMI will release its first AI models quickly, but he's not expecting most people to take notice. The company will first work with partners such as Toyota and Samsung, and then will learn how to apply its technology more broadly. Eventually, he says, AMI intends to develop a "universal world model," which would be the basis for a generally intelligent system that could help companies regardless of what industry they work in. "It's very ambitious," he says with a smile.

UPDATE: Amazon later published a blog post to address what it calls "inaccuracies" in the Financial Times report that the company's own AI tool Kiro caused two outages in an AWS service in December.

An anonymous Slashdot reader had shared this report from the Financial Times: Amazon's ecommerce business has summoned a large group of engineers to a meeting on Tuesday for a "deep dive" into a spate of outages, including incidents tied to the use of AI coding tools. The online retail giant said there had been a "trend of incidents" in recent months, characterized by a "high blast radius" and "Gen-AI assisted changes" among other factors, according to a briefing note for the meeting seen by the FT. Under "contributing factors" the note included "novel GenAI usage for which best practices and safeguards are not yet fully established."

"Folks, as you likely know, the availability of the site and related infrastructure has not been good recently," Dave Treadwell, a senior vice-president at the group, told employees in an email, also seen by the FT. The note ahead of Tuesday's meeting did not specify which particular incidents the group planned to discuss. [...] Treadwell, a former Microsoft engineering executive, told employees that Amazon would focus its weekly "This Week in Stores Tech" (TWiST) meeting on a "deep dive into some of the issues that got us here as well as some short immediate term initiatives" the group hopes will limit future outages.

He asked staff to attend the meeting, which is normally optional. Junior and mid-level engineers will now require more senior engineers to sign off any AI-assisted changes, Treadwell added. Amazon said the review of website availability was "part of normal business" and it aims for continual improvement. "TWiST is our regular weekly operations meeting with a specific group of retail technology leaders and teams where we review operational performance across our store," the company said.

An anonymous reader quotes a report from IEEE Spectrum: Worried that your latest ask to a cloud-based AI reveals a bit too much about you? Want to know your genetic risk of disease without revealing it to the services that compute the answer? There is a way to do computing on encrypted data without ever having it decrypted. It's called fully homomorphic encryption, or FHE. But there's a rather large catch. It can take thousands -- even tens of thousands -- of times longer to compute on today's CPUs and GPUs than simply working with the decrypted data. So universities, startups, and at least one processor giant have been working on specialized chips that could close that gap. Last month at the IEEE International Solid-State Circuits Conference (ISSCC) in San Francisco, Intel demonstrated its answer, Heracles, which sped up FHE computing tasks as much as 5,000-fold compared to a top-of the-line Intel server CPU.

Startups are racing to beat Intel and each other to commercialization. But Sanu Mathew, who leads security circuits research at Intel, believes the CPU giant has a big lead, because its chip can do more computing than any other FHE accelerator yet built. "Heracles is the first hardware that works at scale," he says. The scale is measurable both physically and in compute performance. While other FHE research chips have been in the range of 10 square millimeters or less, Heracles is about 20 times that size and is built using Intel's most advanced, 3-nanometer FinFET technology. And it's flanked inside a liquid-cooled package by two 24-gigabyte high-bandwidth memory chips—a configuration usually seen only in GPUs for training AI.

In terms of scaling compute performance, Heracles showed muscle in live demonstrations at ISSCC. At its heart the demo was a simple private query to a secure server. It simulated a request by a voter to make sure that her ballot had been registered correctly. The state, in this case, has an encrypted database of voters and their votes. To maintain her privacy, the voter would not want to have her ballot information decrypted at any point; so using FHE, she encrypts her ID and vote and sends it to the government database. There, without decrypting it, the system determines if it is a match and returns an encrypted answer, which she then decrypts on her side. On an Intel Xeon server CPU, the process took 15 milliseconds. Heracles did it in 14 microseconds. While that difference isn't something a single human would notice, verifying 100 million voter ballots adds up to more than 17 days of CPU work versus a mere 23 minutes on Heracles.

Last November, Amazon sued Perplexity demanding that the AI search startup stop allowing its AI browser agent, Comet, to make purchases for users online. Today, a judge ruled in favor of the tech giant, granting it a temporary court injunction blocking the scraping of Amazon's website. According to court filings, the judge found strong evidence the tool accessed the retailer's systems "without authorization." CNBC reports: In a ruling dated Monday, U.S. District Judge Maxine Chesney wrote that Amazon has provided "strong evidence" that Perplexity's Comet browser accessed its website at the user's direction, but "without authorization" from the e-commerce giant. Chesney said Amazon submitted "essentially undisputed evidence" that it spent more than $5,000 to respond to the issue, including "numerous hours" where its employees worked to develop tools to block Comet from accessing its private customer tools and to prevent the tool from "future unauthorized access." "Given such evidence, the Court finds Amazon has shown a likelihood of success on the merits of its claim," Chesney wrote.

Chesney's ruling includes a weeklong stay to allow Perplexity to appeal the order. Amazon wrote in its original complaint that Perplexity's agents posed security risks to customer data because they "can act within protected computer systems, including private customer accounts requiring a password." The company also said Perplexity's agents created challenges for the company's advertising business, because when AI systems generate ad traffic, the impressions have to be detected and filtered out before advertisers can be billed. "This requires modifications to Amazon's advertising systems, including developing new detection mechanisms to identify and exclude automated traffic," Amazon wrote in its complaint. "These system adaptations are necessary to maintain contractual obligations with advertisers who pay only for legitimate human impressions."

sziring shares a report from Business Insider: Silicon Valley has long competed for talent with ever-richer pay packages built around salary, bonus, and equity. Now, a fourth line item is creeping into the mix: AI inference. As generative AI tools become embedded in software development, the cost of running the underlying models -- known as inference -- is emerging as a productivity driver and a budget line that finance chiefs can't ignore.

Software engineers and AI researchers inside tech companies have already been jousting for access to GPUs, with this AI compute capacity being carefully parceled out based on which projects are most important. Now, some tech job candidates have begun asking about what AI compute budget they will have access to if they decide to join.

"I am increasingly asked during candidate interviews how much dedicated inference compute they will have to build with Codex," Thibault Sottiaux, engineering lead at OpenAI's Codex, the startup's AI coding service, wrote on X recently. He added that usage per user is growing much faster than overall user growth, a sign that AI compute is becoming even scarcer and more valuable. That scarcity is reshaping how engineers think about their work and pay. "The inference compute available to you is increasingly going to drive overall software productivity," said OpenAI President Greg Brockman.

The report cites a recent compensation submission from a software engineer that listed "Copilot subscription" as part of the pay and benefits. "OpenAI and Anthropic should create recruitment sites where their clients can advertise roles, listing the token budget for the job alongside the salary range," said Peter Gostev, AI capability lead at Arena, a startup that measures the performance of models.

Tomasz Tunguz of Theory Ventures predicts AI inference will be the fourth component of engineering compensation, alongside salary, bonus, and equity. "Will you be paid in tokens? In 2026, you likely will start to be," Tunguz said.

AT&T plans to invest more than $250 billion over the next five years to expand U.S. telecom infrastructure for the AI age. The company says it will also hire thousands of technicians while partnering with AST SpaceMobile to extend coverage to remote areas. Reuters reports: Rapid adoption of artificial intelligence, cloud computing and connected devices has prompted telecom operators to invest heavily in fiber and 5G networks as they also seek to fend off intensifying competition from cable broadband providers. AT&T, which has about 110,000 employees in the U.S., said the new hires will help build and maintain its infrastructure. The outlay includes capital expenditure and other spending, the company said.

The spending will focus on expanding its fiber and wireless networks, including accelerating deployment of fiber broadband, 5G home internet and satellite connectivity to extend coverage across urban, suburban and rural areas. [...] AT&T is also working with satellite partner AST SpaceMobile to expand connectivity to remote regions where traditional network infrastructure is difficult to deploy. The company said it would continue spending on the FirstNet network built for first responders and bolster investment in network security and artificial intelligence-driven threat detection.

OpenAI is reportedly backing away from expanding its AI data center partnership with Oracle because newer generations of Nvidia GPUs may arrive before the facility is even operational. CNBC reports: Artificial intelligence chips are getting upgraded more quickly than data centers can be built, a market reality that exposes a key risk to the AI trade and Oracle's debt-fueled expansion. OpenAI is no longer planning to expand its partnership with Oracle in Abilene, Texas, home to the Stargate data center, because it wants clusters with newer generations of Nvidia graphics processing units, according to a person familiar with the matter.

The current Abilene site is expected to use Nvidia's Blackwell processors, and the power isn't projected to come online for a year. By then, OpenAI is hoping to have expanded access to Nvidia's next-generation chips in bigger clusters elsewhere, said the person, who asked not to be named due to confidentiality. In a post on X, Oracle called the reports "false and incorrect." However, it only said existing projects are on track and didn't address expansion plans.

CNBC notes: "Oracle secured the site, ordered the hardware, and spent billions of dollars on construction and staff, with the expectation of going bigger."

An anonymous reader quotes a report from The Register: AI can reverse engineer machine code and find vulnerabilities in ancient legacy architectures, says Microsoft Azure CTO Mark Russinovich, who used his own Apple II code from 40 years ago as an example. Russinovich wrote: "We are entering an era of automated, AI-accelerated vulnerability discovery that will be leveraged by both defenders and attackers."

In May 1986, Russinovich wrote a utility called Enhancer for the Apple II personal computer. The utility, written in 6502 machine language, added the ability to use a variable or BASIC expression for the destination of a GOTO, GOSUB, or RESTORE command, whereas without modification Applesoft BASIC would only accept a line number. Russinovich had Claude Opus 4.6, released early last month, look over the code. It decompiled the machine language and found several security issues, including a case of "silent incorrect behavior" where, if the destination line was not found, the program would set the pointer to the following line or past the end of the program, instead of reporting an error. The fix would be to check the carry flag, which is set if the line is not found, and branch to an error.

The existence of the vulnerability in Apple II type-in code has only amusement value, but the ability of AI to decompile embedded code and find vulnerabilities is a concern. "Billions of legacy microcontrollers exist globally, many likely running fragile or poorly audited firmware like this," said one comment to Russinovich's post.