Anthropic’s Mythos offline as talks drag for weeks

Anthropic’s Mythos – Two weeks after the Trump administration ordered Anthropic to suspend Mythos-class models tied to export controls, Anthropic has kept the systems offline with no clear end date. Behind the silence is a knot of security rules, a fast-moving approval pathway tha

For two straight weeks. Anthropic’s Mythos-class models have stayed offline—pulled out of circulation after a Friday evening ultimatum from the Trump administration. The company moved quickly, dispatching executives to Washington, DC. But the weeks since have brought something more unsettling than delay: almost no clarity.

Anthropic declined to comment multiple times this week about the state of the talks. saying there was no news to share. Yet with 14 days of high-intensity negotiations behind it. nobody outside the negotiating rooms can say when. or whether. Anthropic’s most powerful AI models will come back online. And if the administration’s order expands beyond Anthropic—potentially to more companies with similar technology—the disruption would ripple far past one startup.

The trigger is direct and specific. The Trump administration’s June 12 export control order demanded that Anthropic suspend access by “any foreign national” to Mythos 5 and Fable 5 due to security concerns. The ban covered any non-US citizen inside or outside the US, including people employed by Anthropic. So far, Anthropic has concluded that its only option is to keep these models offline.

What’s harder to pin down is why the talks have stalled. One key problem. as the model of export controls collides with the realities of AI. is that there’s no clear framework for applying export controls to AI systems. Most companies that build dual-use products—civilian systems with potential defense or military applications—can often evaluate risk using what amounts to a checklist during manufacturing and production. Anthropic, in contrast, is stuck navigating a complicated bureaucracy trying to apply the rules from first principles.

That process. according to how export controls typically play out. can take months or years and finish before a product reaches market. But here the timeline appears to have compressed sharply. As previously reported, the US Department of Commerce apparently tested Fable 5 before release and raised no complaints. A source familiar with negotiations said Anthropic concluded its models were safe to release. The agency apparently did not act until someone—reportedly Amazon CEO Andy Jassy—flagged a method for seemingly breaking Fable 5’s guardrails. Then, the entire process was pushed into a few days.

Katie Moussouris. founder and CEO of Luta Security. weighed in after Anthropic asked her to view a report about the Fable 5 vulnerability. In her view, the concern is significantly overblown. In a blog post. Moussouris described how researchers jailbroke guardrails that prevent Fable 5 from finding exploitable security holes—capabilities she calls among the “unfettered Mythos 5’s scariest capabilities.” She says the model would refuse requests to review code “for security issues. ” but it would accept demands to “fix this code” followed by manual prompts. In theory, that chain could lead the model to flag vulnerabilities it wasn’t supposed to divulge.

Moussouris argues that this still shouldn’t have triggered the scale of government action. She portrays the technique as essential to AI coding—particularly for defensive security work. “Defenders need to be able to ask AI to fix the bugs in a file. explain why the fix matters. and write tests that confirm the patch works. ” she wrote. “That is not a guardrail bypass. It is the most valuable thing an AI model can do for defensive security: executing the find. fix. and test loop defenders run every day.”.

Inside Anthropic’s negotiation team, there has been movement. In the past week, cofounder Tom Brown replaced CEO Dario Amodei in negotiations with the Trump administration, alongside Sarah Heck, the company’s public policy chief. But even with those shifts, progress still appears slow.

The human stakes inside the company are obvious in the way the business was structured to depend on Mythos. Before the negotiations began, Anthropic was seen as one of the rare AI companies with a path to profitability. Its Mythos-class models were designed to boost revenue ahead of an upcoming IPO. Mythos’ input tokens sell for double the cost of the lower-powered Opus 4.8. and Mythos’ cybersecurity prowess had even seemed to thaw relations with the Trump administration after months of legal and rhetorical combat.

Anthropic now needs Mythos revenue to cover compute expenses it has secured recently. That includes a deal to pay SpaceX $15 billion per year for access to its data centers. as well as its public image before the IPO. Two of Anthropic’s largest current shareholders—Google and Amazon—have tried to carefully stay on Trump’s good side. meaning they are likely not pleased by the uncertainty as models remain unavailable.

The broader market feels the pressure too. With Mythos shut down and export controls creeping toward other American systems. the US AI sector is suddenly operating with a power vacuum it didn’t ask for. The US government has signaled a willingness to lock down American AI systems it deems risky. and several US companies—including OpenAI. Google. and Microsoft—have models that may pose similar risks to Mythos. At the same time, countries have started calling for non-American AI, pushing the race into a more international political frame.

Alex Stamos, cybersecurity expert and chief product officer at Corridor, described the moment in blunt terms: “One of America’s champions is being kneecapped by the US government while we’re in a race with the Chinese. It’s just incredibly stupid.”

The timing only tightens the scenario. As days pass, more companies edge toward Mythos-level capabilities that could trigger their own export control orders. OpenAI’s GPT-5.5 Cyber reportedly beat Mythos 5 on certain benchmarks. The Trump administration has also reportedly asked OpenAI to delay the release of GPT-5.6 over security concerns. with plans for the government to approve each customer one by one. Anthropic and OpenAI’s IPOs are both approaching, and every day China pulls further ahead in the AI race.

There’s also a sharp irony in how this unfolded. The Trump administration’s June 12 export control order arrives after months of pushing to dismantle AI safeguards and regulation. It is one of the first sweeping regulatory decisions President Trump has made. Yet cybersecurity leaders have been pushing back collectively. saying if regulation has to happen. this is not the right way to do it. For all of the administration’s vows to roll back Biden-era AI regulation. it has gained that ground back—and then some. not through easing rules. but through decisions that keep leading models offline and draw new boundaries around who can access them.

In Washington, Anthropic’s executives keep meeting the government’s moving target. In the company’s product world, Mythos remains dark. And in the wider AI market, uncertainty spreads faster than any rollout could ever fix.

Anthropic Mythos 5 Fable 5 export controls AI regulation Trump administration Tom Brown Dario Amodei Sarah Heck cybersecurity Luta Security Katie Moussouris SpaceX data centers IPO GPT-5.5 Cyber GPT-5.6