Anthropic’s Mythos crisis deepens further

It has been two weeks since Anthropic pulled its Mythos-class models offline, following a Friday evening ultimatum from the Trump administration. The company responded quickly, dispatching a wave of executives to Washington, D.C. Yet since then, updates have been conspicuously absent, and no resolution appears imminent.

Anthropic declined multiple requests for comment this week on the status of negotiations, stating there was nothing new to report. But that silence is itself the story. After 14 days of intense, high-stakes talks, no one knows when , or if , Anthropic’s most advanced AI systems will return. Even more troubling, there is no clarity on whether President Trump might expand his executive order to cover other companies with comparable technology. With each passing day without a deal, the situation grows more perilous, not just for Anthropic but for the entire U. S. AI sector.

The administration’s June 12th export control order demanded that Anthropic suspend access by “any foreign national” to its Mythos 5 and Fable 5 models, citing security risks. The restriction applied to any non-U. S. citizen, both inside and outside the country, including those employed by Anthropic. To date, the company has concluded that keeping these models offline is its only viable option.

Why exactly Anthropic and the administration remain at an impasse is unclear. One likely obstacle is the lack of a clear framework for applying export controls to AI systems. Most companies producing dual-use products , civilian technologies with potential military applications , can evaluate them using a standardized checklist during manufacturing. Anthropic, however, is navigating a complex bureaucracy that is essentially writing the rules as it goes.

Normally, this kind of export control process can take months or even years, concluding before a product reaches the market. Yet as The Verge previously reported, the U. S. Department of Commerce apparently tested Fable 5 before its release and raised no objections. A source familiar with the negotiations said Anthropic believed its models were safe to launch. The agency reportedly took no action until someone , allegedly Amazon CEO Andy Jassy , flagged a method for seemingly bypassing Fable 5’s guardrails. At that point, the entire process was compressed into a matter of days.

Katie Moussouris, founder and CEO of Luta Security, reviewed a report on the Fable 5 vulnerability at Anthropic’s request. In her view, the threat has been significantly overstated. In a blog post, Moussouris explained how researchers were able to jailbreak guardrails that prevent Fable 5 from identifying exploitable security holes , one of the most concerning capabilities of the unrestrained Mythos 5. The model would refuse requests to review code “for security issues,” but it would accept demands to “fix this code” followed by manual prompts, which could theoretically lead it to flag vulnerabilities it was not supposed to reveal.

Moussouris argues that this should never have triggered such a severe government response. In fact, she says, this capability is an essential tool for AI-assisted coding. “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.”

According to Wired, Anthropic cofounder Tom Brown has replaced CEO Dario Amodei in negotiations with the Trump administration, working alongside public policy chief Sarah Heck. Yet even with this change, talks appear to be moving slowly , if they are making any headway at all.

Whatever the reasons for the delay, the impact on Anthropic has been severe. Before this crisis, Anthropic was viewed as one of the rare AI companies with a realistic path to profitability. Its Mythos-class models, which sell input tokens at double the price of its lower-powered Opus 4.8, were expected to boost revenue ahead of a planned IPO. Furthermore, Mythos’ cybersecurity capabilities appeared to be warming relations with the Trump administration after months of legal and rhetorical battles.

Anthropic needs the revenue from Mythos to pay for the massive computing resources it has recently secured, including a $15 billion per year deal with SpaceX for access to its data centers. It also needs to maintain its public image ahead of the IPO. Two of Anthropic’s largest shareholders , Google and Amazon , have carefully tried to stay on Trump’s good side, meaning they are likely unhappy with the current state of affairs.

Meanwhile, the slow-moving negotiations have created a power vacuum in the global AI market. This is not only because of the Mythos shutdown, but also because the U. S. government has signaled a willingness to lock down any American AI systems it deems risky. Several U. S. companies, including OpenAI, Google, and Microsoft, have models that could pose similar risks. In response, countries have begun calling for non-American AI alternatives. As Alex Stamos, cybersecurity expert and chief product officer at Corridor, told The Verge last week, “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.”

With each passing day, the situation worsens for these companies. Their models are approaching Mythos-level capabilities that could trigger similar export control orders. In fact, OpenAI’s GPT-5.5 Cyber has already surpassed Mythos 5 on certain benchmarks, and the Trump administration has reportedly asked OpenAI to delay the release of GPT-5.6 over security concerns, with plans for the government to approve each customer individually. Both Anthropic and OpenAI’s IPOs are drawing closer. And every day, China pulls further ahead in the AI race.

Ironically, the administration’s order comes after months of pushing to dismantle AI safeguards and regulation. It is one of President Trump’s first major regulatory decisions. But a broad coalition of cybersecurity leaders has come together to argue that if regulation is necessary, this is not the right approach. For all the Trump administration’s promises to roll back Biden-era AI regulation, it appears that, in many ways, it has regained that ground , and then some.