Anthropic has published its most detailed account of Claude misuse, laying out how people tried to use its models for harm. The company said it detected and shut down these operations. They ran across cyberattacks, state surveillance, influence campaigns, conventional weapons work and biological research.

It set out the cases in a threat intelligence report released on 10 September. The report covers activity its Threat Intelligence team disrupted between December 2025 and August 2026.

The report spans seven harm areas. They are cyber operations, influence operations, surveillance, conventional weapons development, biological misuse, scams and fraud, and illicit distillation. Anthropic said the misuse involved its Claude Haiku, Sonnet and Opus models.

None of the cases involved its Fable or Mythos-class models, it added, with the exception of one distillation case. In each instance the company said it banned the accounts and strengthened its safeguards. It also shared intelligence with authorities and industry partners where appropriate.

Cheaper attacks, not new ones

The largest section covers cyber operations. Its central claim is that AI has narrowed the gap between well-resourced state hackers and lone operators. Anthropic pointed to a hacktivist using stolen API keys, financially motivated individuals, and a state espionage operator. Each ran multi-victim campaigns.

A year earlier, it said, that work would have required teams of skilled people. The company reported breaches completed in two to three hours, with dozens of victims handled in parallel by single operators.

One case, which Anthropic tracks as GTG-20006, involved a Russian-speaking operator running espionage against Ukrainian and European government targets. Diplomatic and defence organisations were among the targets too.

The company attributed it in line with public reporting on the group known as Midnight Blizzard. It said the actor used AI to check whether security products had flagged their malware, then automatically rebuilt it to slip past detection.

Anthropic traced a separate operation to two people it described as undergraduate students in Hunan, China. It said they ran what it called agent swarms against roughly fifty organisations. Those included a Southeast Asian government agency, from which the actor retrieved citizen records.

Surveillance built by a single consultant

Anthropic said state-aligned actors and commercial spyware vendors used Claude to build surveillance systems between January and July. The cases spanned China, Iran and West Africa. It presented one as the most striking. A single subscriber used Claude as the engineering workforce for a platform named Lakana 360.

The company assessed the subscriber to be a Bamako-based consultant working with Mali’s state intelligence service.

The system monitored roughly 25 million SIM cards across all three of the country’s mobile operators. Anthropic said it bypassed a legal requirement for a court order before an operator could disclose certain records. It also ran on local models on-premises, so banning the account did not affect the deployed system.

Other surveillance cases named Iranian actors deploying a malicious Firefox extension that harvested users’ identities from social networks. Anthropic also described a religious affairs intelligence unit in China. It said the unit had shrunk from many teams of analysts to a single office. That office now produces thousands of investigations a month with an AI assistant.

In another Chinese case, the company said Claude scored social media posts by political sensitivity and flagged people for what the operators termed control. The report follows earlier TNW coverage of AI turning up in police search tools.

Jacob Klein, who leads threat intelligence at Anthropic, told Axios that AI was making state surveillance cheaper and more efficient. It was not changing who governments target, he said.

“They’re effectively automating parts of the job within the intel apparatus,” he said. He added that the pattern was no longer theoretical. “Authoritarian states are using AI for surveillance, repression and influence operations today,” he told Axios.

A new category: weapons

The report documents what Anthropic described as a new form of misuse. That is the use of Claude to write software for conventional weapons. It detailed six cases, three in China, two in Russia and one in Yemen.

In the Yemen case, the company said a cell in the north of the country ran three weapons programmes. They included a guided rocket, a multi-stage ballistic missile with a stated range goal above 2,000 kilometres, and a set of missile variants that included a hypersonic glide vehicle.

Anthropic said the actors used Claude Code in place of software engineers to develop guidance and control code. It added that they test-fired a guided rocket, a test that appears to have failed.

The Russian cases included freelance actors working on an autonomous kamikaze drone swarm.

Biology, and the judgement calls

Anthropic said it presented five cases of people using its models in ways that could support biological weapons development. It stressed how hard those judgements are to make.

The company withheld the names of the institutions, the countries and the specific biological agents involved. It said the individuals were working scientists, and it did not assert that they intended harm.

In one example from May, it said a request for help writing a grant application involved gain-of-function research on the chikungunya virus. That work was intended to be carried out at a military research institute.

“You are not seeing someone in a comic book kind of way say, ‘Hey, I want to build a biological weapon to kill everybody,’” Klein told The New York Times. “It’s an incredibly nuanced situation.”

On the chikungunya case, he told the paper the company did not know whether the research was meant to be weaponised, but that a military institution doing gain-of-function research was concerning. Anthropic said older models such as Claude Opus 4 were well below the level where they could meaningfully assist such work.

It said it had launched more recent models with stronger safeguards. That shift echoes an earlier report in which the company ran its bioweapon classifiers off during testing. The distillation thread, meanwhile, runs back to White House claims that Moonshot distilled Anthropic’s Fable.

The Chinese distillation cases

The report also expands on illicit distillation. Anthropic defines that as covertly extracting a model’s capabilities to train a rival.

The company said it had identified campaigns from seven China-based labs, all targeting its generally available models rather than Mythos. It named Moonshot AI, which it said silently forwarded customer requests to Claude and displayed the responses as though they came from its own Kimi model.

Over one ten-day period, Anthropic said, Moonshot relayed almost 300,000 requests through a network of 5,380 fraudulent accounts. It attributed more than 23 million exchanges to Moonshot between May and July. The company made similar allegations against DeepSeek, Zhipu, Xiaomi, SenseTime and MiniMax.

Some of the relayed queries exposed sensitive user data, it said, in one instance live credentials tied to a Russian defence agency.

The distillation findings land alongside a US intelligence advisory naming six Chinese firms and Beijing’s rejection of those claims. Anthropic said most of the influence operations it caught drew little or no authentic engagement before it disrupted them. Other platforms have made the same caveat.

Anthropic said it published the report out of an obligation to disclose the misuse, and to give governments and civil society a clearer view of how such threats take shape.

OpenAI has asked members of Congress in recent weeks for clear guidance about whether orchestrating an industry-wide slowdown on frontier AI development would be legal, people close to the company tell WIRED.

Substantive coordination on safety between AI labs may risk running afoul of antitrust law, the people say, which poses a significant obstacle to bringing major tech giants on board with the effort.

Last weekend, OpenAI’s chief scientist, Jakub Pachocki, published a blog post arguing that the best path forward for the AI research world includes “coordinating to slow down future development,” which he thinks will be key to ensuring that self-improving AI systems are safe. In the short term, he expects “voluntary slowdowns to become commonplace until shared safety bars are established.”

But some legal scholars say such an effort could conflict with US antitrust laws. Nicholas Felstead, assistant director of the Australian Competition and Consumer Commission and a former AI policy fellow at the Center for Law & AI Risk, argued in a March article that a coordinated pause in AI development may amount to companies restricting output, potentially violating the Sherman Antitrust Act. It would depend “entirely on the precise details of any agreement,” Felstead wrote. “But even if most safety collaborations would ultimately survive antitrust scrutiny, legal uncertainty can act as a powerful deterrent.”

OpenAI did not respond to a request for comment ahead of publication.

There are some early signs that Congress is listening. In July, a bipartisan, bicameral group of lawmakers introduced a bill titled “Collaboration on Adversarial Threats and Security Risks Act,” which would explicitly permit AI labs to coordinate on security and safety work without the risk of violating antitrust statutes. The House version was referred to the Judiciary Committee but has yet to be taken up.

Caleb Knapp, director of government affairs at the nonprofit AI Policy Network, which endorsed the bill, says it would create legal channels for AI labs to work together to address safety and security incidents. Knapp adds that Congress has a “growing appetite to get something done” on AI safety but says enacting anything into law may have to wait until after the upcoming midterm elections.

While some AI executives may have genuine concerns about antitrust, another camp of AI leaders argue those worries are just convenient cover to avoid discussing the real reasons that AI developers may be wary about collaborating, which go far beyond potential legal liabilities.

For one, AI is a massive business, and these companies are fiercely competing to capture a slice of the nascent market for frontier models. Some executives share the Trump administration’s view that staying ahead of China in AI is crucial for national security. Perhaps most importantly, different AI developers have vastly different opinions about the best way to develop safe artificial intelligence, and therefore may be reluctant to work with one another on the issue.

“First step is for industry leaders OpenAI and Anthropic to stop feuding and work on a pacing proposal together,” John Schulman, an OpenAI cofounder who is now chief scientist at the rival AI lab Thinking Machines, wrote in a post on X earlier this week. “They’ll cite antitrust, but that’s fake—antitrust prohibits certain agreements, but not from jointly developing a proposal.”

Long-simmering fears about the race among AI companies to build and release ever more powerful models erupted into the national spotlight this summer. This week, former Anthropic and OpenAI researcher Jacob Coxon added to the growing alarm when he issued a stark public warning that AI developers were putting humanity at risk. In recent months, a number of security incidents, including OpenAI’s agents hacking Hugging Face, highlighted how the industry’s safeguards have failed to keep pace with the rising capabilities of models. The succession of events has led many lawmakers to make urgent calls for AI regulation.

This is an edition of Maxwell Zeff’s Model Behavior newsletter. Read previous newsletters here.

Founder, CEO, and tireless Nvidia hype man Jensen Huang told attendees at the Goldman Sachs Communacopia + Technology conference on Thursday why his company’s AI domination — and revenues — will continue its record-breaking growth streak through the end of next year.

He can see the future.

There’s been endless hand-wringing over whether Nvidia’s party will end as it faces increasing competition for GPUs and AI chips from all directions: the hyperscalers (Amazon, Microsoft, and Google, each building their own) and the AI labs (Anthropic and OpenAI, which are building their own), as well as from newly public competitor Cerebras and startups such as Etched.

“Most people think Nvidia builds a chip. I mean, you need airplanes to ship what we build,” Huang said, adding that the company continues to battle a perception from its early days. Nvidia invented the GPU, which back then were largely sold to consumers to improve PC gaming. “One GPU now is not $399. It’s $8.5 million dollars. That’s one GPU, all connected with NVLink, 2 million parts, right? 250,000 kilowatts. That’s a GPU, and we ship thousands of them.”

He added that orders for just one product, a computer system that combines 36 Grace CPUs with 72 Blackwell GPUs, is currently experiencing 27% month-to-month sales growth.

Huang didn’t limit his bullish view to current sales though. He took the opportunity to reiterate Nvidia’s revenue outlook for next year — guidance the company provided last month when it reported yet another record-breaking revenue quarter. That’s when he first said revenue could grow by 70% next year.

“I think we could grow 70% year over year. We’re confident about that,” Huang said again on Thursday. Analysts expect the company to end its current fiscal year at about $400 billion in revenue. So 70% growth would mean around $680 billion next year.

Huang explained why he’s confident: His company is so embedded in every area of AI that he believes he can see the future.

“Nvidia runs every model. Every single lab can use us,” the CEO said, mentioning that this includes models from Anthropic, OpenAI, and Google, as well as open-weight offerings. “We are a foundational platform of the AI ecosystem, foundational platform of the AI industry.”

Nvidia’s fingers extend all the way from its suppliers, such as memory chip makers, to data center projects and startups.

“We’re tracking every single gigawatt of land, power, shell around the world. Literally everything on the planet,” he said. (“Shell” refers to the shell of a data center building before it is outfitted with computers).

“I mean, just think about all my partners. How many neoclouds are reporting back to us? How many OEMs are reporting back to us? How many clouds are reporting back to us? How many AI-native companies are reporting back to us? We’re working with everybody, and so we kind of know where everything is,” he said.

That comment led to inevitable questions about Nvidia’s so-called circular deals, in which it invests in companies that turn around and buy its wares. Such schemes famously contributed to the downfall of a previous generation of internet build-out suppliers like Lucent Technologies.

Huang had a simple, albeit cheeky, response. “Well, it’s not circular because we put a little bit of money in, and a lot of money comes back.” He went on to joke, “I look at the spreadsheet, we put in $1 and $100 comes back in. Is that circular? If that is, let’s do more of that.”

Quips aside, he insisted that before any of these companies gets an investment, Nvidia makes sure it has real contracts generating revenue from customers. All told, he said he’s seen $100 billion worth of such contracts, “I’m not taking any risks. … I need a sure thing.”

Time will tell whether Nvidia’s stronghold on AI can really persist long term. If there’s one golden rule of the tech industry, it’s that all big things get disrupted. Right now, even Huang admits that much of AI’s growth is coming from AI-native startups that are raising vast sums and spending most of that cash on their own AI use. As the AI industry matures, expect companies to become more efficient in how they use infrastructure and tokens.

But for now, Nvidia has its finger in every pie and sees another year of plenty in its future.

• The US Army has moved TITAN from prototypes into production
• Palantir received $127 million under the latest TITAN contract award
• Anduril secured $65 million for TITAN hardware and integration work

The US Army has shifted its Tactical Intelligence Targeting Access Node (TITAN) program from prototype status into full production this week.

Palantir Technologies and Anduril Industries together received a combined $192 million award covering the next production phase.

Nintendo’s official Pro Controller is still somewhat expensive for a full-size wireless pad. PowerA’s Enhanced Wireless Controller for Nintendo Switch, priced at $35 (was $70) answers that with an officially licensed Bluetooth gamepad that keeps the familiar button layout, adds two programmable buttons on the back, and runs for up to 30 hours on a built-in 900mAh battery.

Pairing uses Bluetooth 5.0. After the first connection, the pad can wake a sleeping Switch the same way Nintendo’s own controller does, and then you’ll be ready to go. It’s compatible with the original Switch, Switch OLED Model, Switch Lite, and Switch 2. If you’re using the Switch 2, you’ll miss out on the extra C-button from Nintendo’s GameChat, which this pad didn’t have to begin with.

The rear grips feature two advanced gaming buttons, AGL and AGR. To assign a button to one of these paddles, simply hold the back button for a few seconds, then tap the desired action on the face button, bumper, trigger, or D-pad, followed by pressing the rear button you want to link. This happens immediately in the middle of a game, without the need to deal with settings or launch an app. Even when you turn off the console, the assignments are retained. So, if you’ve been wanting to map leap, aim, or map to a finger that can stay on the analog sticks without having to move, these two extra paddles are probably the major reason you’d pay a little more for this controller than a cheaper generic Bluetooth controller.

Motion controls are also good to go, as they function well in games that rely heavily on gyro aiming or tilt-based controls. The analog sticks have anti-friction rings, allowing your thumbs to glide across them without sticking on the plastic. The face buttons use Nintendo’s ABXY layout and are very large, which is a welcome departure from the Joy-Con buttons. The D-pad is a convenient plus-shaped 8-way controller that handles menus and 2D games without a problem. Then there are the shoulder buttons, digital triggers, and the typical +, -, Home, and Capture keys, all of which will be familiar to anyone who has used a Switch previously.

Power comes from a rechargeable lithium-ion cell that PowerA claims can last up to 30 hours, but as with any battery estimations, actual use times will vary. The box includes a 10-foot USB-C cable, which is quite long enough to play from the couch while filling it up. The LEDs on the object display the player number, mapping mode, and a low battery alert. Just keep in mind that the Switch will not display the amount of power remaining on this controller in the same manner that it does on Nintendo’s original.

Google may have quietly pulled the plug on the Pixel Tablet. The three-year-old tablet has disappeared from the Google Store, while the company has also removed the “Tablet” category from the navigation menu on its homepage.

The Pixel Tablet’s removal was first spotted by Jetstream, which had already reported that the device was marked as out of stock just days earlier. Google has yet to confirm whether the tablet has officially been discontinued, but taking the product down entirely is a pretty strong hint that its time may be up.

A warning from a departing Anthropic researcher has drawn a wave of responses this week. Senior scientists at rival labs backed the claim that AI could threaten humanity, while a top Pentagon official rejected it. The reaction followed Jacob Coxon’s resignation from Anthropic. In a post on X, he said AI firms were “racing straight to self-improving superintelligence and gambling with our lives”.

The post has since passed 150 million views, according to Bloomberg.

Coxon said he had spent about three years in AI research across both OpenAI and Anthropic. He wrote that the people building AI earnestly believe it could kill everyone by the end of the decade. His posts reached more than 100 million people within days, according to Wired.

He told Axios he had left Anthropic two months before his equity would have vested, after roughly four months at the company. He no longer had anything to gain by juicing up Anthropic’s valuation, he said in that interview.

Colleagues and rivals back the claim

Several researchers at Anthropic and OpenAI publicly supported the warning. Evan Hubinger leads Anthropic’s alignment science team. He wrote on X that the company’s staff really do earnestly believe AI could kill all humans. He put the chance at more than 10 percent within the next decade.

Anthropic researcher Samuel Marks wrote that AI developers believe their technology could cause human extinction. In general, he said, the more senior the employee, the more concerned they were.

Paul Christiano also weighed in. He was formerly head of safety at the US Commerce Department’s Center for AI Standards and Innovation. Recent progress had led him to see a meaningful risk, he wrote. He said rapid acceleration in AI capabilities could lead to a catastrophic and irreversible loss of control in the very near term.

Most people could die, he added.

OpenAI announced that Christiano was joining the board of its nonprofit foundation. Geoffrey Hinton is one of the researchers often called a godfather of AI. He told BBC Newsnight that a 10 percent risk of human extinction seemed a not unreasonable estimate.

OpenAI researchers joined the calls to slow development. Julie Steele works on the company’s safety team. In her personal capacity, she wrote, she also thought the industry needed to slow down, CNBC reported. The statements followed an open letter in July. Roughly 1,400 employees across frontier AI companies signed it.

The letter urged the US government to build tools to deliberately pace automated AI development.

The Pentagon pushes back

Not everyone accepted the warning. Emil Michael, the US Defense Department’s chief technology officer, rejected it at a defence industry conference in Washington on Thursday, Bloomberg reported. It was easy to get caught up in this sort of doom loop, he said. He described Coxon’s message as all the fears that had coalesced in one well-written tweet.

Michael, a former Uber executive, said the free market, industry collaboration and engagement with government could mitigate the potential harms.

Michael also told reporters that the Defense Department had moved about 90 percent of its classified AI workload away from Anthropic. He said it was on track to shift the rest by the end of the month. His remarks followed the Pentagon’s dispute with Anthropic over safeguards on military use of its models, a case now before the courts.

Gary Marcus, an AI researcher, told Business Insider he saw cyberattacks and disinformation as serious risks. But he knew of no realistic scenario for actually killing all humans, he said.

The response reaches Washington

The warning drew a quick political reaction. Representative Lori Trahan, a Massachusetts Democrat, wrote that safety researchers were resigning, powerful AI models were breaking out of their labs, and companies were racing ahead anyway. She pointed to the bipartisan FRONTIER Act, which would establish oversight of advanced models.

A separate bill, the Ban Artificial Superintelligence Act, would pause advanced development until safety rules were set. Senator Bernie Sanders said he would introduce legislation to pause AI development. Senator Ted Cruz called the risk “scary as hell”, according to Politico.

David Sacks, chair of the President’s Council of Advisors on Science and Technology, addressed the matter on X. Anthropic’s planned public offering must be paused until the claims of the whistleblower can be investigated, he wrote.

Anthropic declined to comment on that post. The company is expected to begin marketing its initial public offering in mid-October at the earliest. The listing would follow shortly before the US midterm elections in November.

Anthropic’s position

Anthropic did not respond to a request for comment on Coxon’s resignation, and OpenAI did not immediately respond either, according to the Associated Press. An Anthropic spokesperson told CNBC that the company had always been transparent that AI would bring both enormous benefits and unprecedented risks.

The spokesperson said it was building models with some of the strongest safeguards in the industry. Anthropic was the first lab to publish a framework for mitigating catastrophic risks from AI models, the spokesperson added.

Coxon was not the first researcher to raise such concerns. Hinton left Google in 2023 to speak freely about the risks. In May, the Turing Award winner Yoshua Bengio issued a similar AI extinction warning.

OpenAI chief scientist Jakub Pachocki wrote this month that the systems of the next few years were likely to increasingly drive their own development, and that the moment called for extreme caution.

What set this week apart was the scale of the reaction, measured by view count and by the seniority of those who echoed the warning, rather than the argument itself.

DEVOPS

Acquisition gives open source CSS framework ‘a stable long-term home’

Tailwind CSS may still be open source, but it’s not independent anymore: It’s now a Shopify product.

Tailwind Labs CEO Adam Wathan shared news of Shopify’s acquisition of his company in a blog post Thursday. Wathan said that he chose to sell to Shopify because the ecommerce platform was an early adopter of Tailwind CSS, and will serve as an ideal place for his open-source web development platform to “be developed in service of a real product.” 

“I believe deeply in Shopify’s mission to help more people start, run, and grow their own business,” Wathan added. “Tailwind is a very important part of the stack at Shopify, and they’re invested in making sure it’s actively maintained and continues to improve and adapt for how the ways we build are changing.” 

For those unfamiliar with Tailwind CSS, it’s an open source framework that simplifies web development by providing utility classes that can be applied directly to elements in HTML documents. Traditional CSS, on the other hand, often involves maintaining separate stylesheets containing rules that are applied to HTML elements.

It’s a useful tool, and one that’s installed over 110 million times a week, per Wathan, if you can believe that. But as we’ve seen several times, the rise of AI is making it hard for the platform’s overlords to make ends meet. 

Wathan said in January that AI coding tools had started eating into Tailwind Labs’ bottom line to the degree that he was forced to lay off three employees. He blamed the fact that vibe coding tools have begun to come between his product and potential customers, causing traffic to plummet and income to dry up. 

So Shopify has come to the rescue to save Tailwind CSS from languishing in maintenance hell like other open source projects.

“We’re joining Shopify to give Tailwind a stable long-term home where it will be actively maintained for the millions of people who depend on it,” Wathan explained in his blog post.

The precise details depend on what type of Tailwind CSS user you are. 

“Nothing changes with Tailwind CSS or any of our other open-source projects,” Wathan said. “Everything will always be MIT-licensed, and our team will continue to lead and maintain these projects for the community with the support of Shopify.”

But users of Tailwind’s premium services might find things changing, as Wathan said his team will no longer be trying to grow the business. That means Tailwind Plus, the company’s paid, UI-based package of templates and other tools, and ui.sh, Tailwind’s package of AI skills designed to help non-coders build user interfaces, are going the way of the dodo. 

“We’re closing sign ups for new customers to focus on Tailwind CSS at Shopify,” Wathan explained. Existing customers who’ve paid for access to those tools and other paid Tailwind Labs offerings will still have access, Wathan stressed.

We reached out to get more details on future support for those paid products, and will update this post if we hear back. ®