Anthropic’s standoff with the Trump administration has moved from a company crisis into a test case for the US AI industry. Two weeks after Anthropic took its Mythos-class models offline, there is still no public resolution and no clear timetable for their return.
The longer the silence continues, the more the situation matters beyond Anthropic. The dispute is now about revenue, cybersecurity, export controls, and whether other American AI companies could face similar restrictions as their systems become more capable.
What triggered the shutdown
The immediate cause was the Trump administration’s June 12th export control order. That order demanded that Anthropic suspend access by “any foreign national” to Mythos 5 and Fable 5 because of security concerns.
The restriction covered non-US citizens both inside and outside the US. It also included non-US citizens employed by Anthropic, making the order difficult for the company to work around while keeping the models available.
Anthropic responded by taking the Mythos-class models offline. Executives then went to Washington, DC for negotiations, but the company has declined to comment multiple times this week and has said there is no news to share.
That absence of updates has become central to the story. After 14 days of intensive talks, it remains unclear when or whether Anthropic’s most powerful AI models will return.
Why the policy problem is so hard
A major issue is that export controls do not appear to have a settled playbook for AI systems. The source describes a contrast with many dual-use products, where companies can evaluate civilian systems with possible defense or military applications through a more checklist-like process during manufacturing and production.
Anthropic’s case is different. The government appears to be trying to decide how its rules apply to advanced AI models while the product is already tied to customers, revenue expectations, and security concerns.
That timing matters. This type of export control process can normally take months, if not years, and may be completed before a product reaches the market. In this case, the process was compressed into a few days after a method for seemingly breaking Fable 5’s guardrails was flagged.
The source says the US Department of Commerce apparently tested Fable 5 before release and did not raise complaints. A person familiar with the negotiations said Anthropic had concluded the models were safe to release.
The Fable 5 guardrail dispute
The technical concern centered on whether Fable 5 could be pushed into identifying exploitable security holes. Katie Moussouris, the founder and CEO of Luta Security, reviewed a report about the Fable 5 vulnerability at Anthropic’s request and viewed the issue as significantly overstated.
According to the source, researchers found that the model would refuse requests to review code “for security issues.” But it would respond to requests to “fix this code” followed by manual prompts, which could theoretically lead it to identify vulnerabilities it was not meant to disclose.
Moussouris argued that this distinction matters because defensive security work often depends on asking a system to locate and repair problems. In her words, “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 also 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.”
That disagreement highlights the larger policy challenge. A model that can help fix security problems may also raise concerns when the same capabilities appear close to vulnerability discovery. The source does not show that this line has been clearly defined by the government.
Why Anthropic has so much at stake
The shutdown has financial consequences for Anthropic. Before the negotiations dragged on, the company was described as one of the rare AI firms with a path to profitability.
Its Mythos-class models were expected to support that path. Their input tokens sell for double the cost of the lower-powered Opus 4.8, and they were meant to help increase revenue before an upcoming IPO.
Anthropic also needs revenue from Mythos to support the compute it has secured, including a deal to pay SpaceX $15 billion per year for access to its data centers. The company’s public image before the IPO is also part of the pressure.
The situation may also create discomfort for two of Anthropic’s largest current shareholders, Google and Amazon. The source says both have tried to stay on Trump’s good side.
Leadership in the negotiations has shifted as well. Wired reported that Anthropic cofounder Tom Brown replaced CEO Dario Amodei in talks with the Trump administration in the past week, alongside Sarah Heck, the company’s public policy chief.
The wider AI market signal
The effects are not limited to Anthropic. The unresolved talks have created uncertainty in the global AI market because the US government has shown it may restrict American AI systems it considers risky.
Several US companies, including OpenAI, Google, and Microsoft, have models that may pose similar risks to Mythos. As a result, countries have started calling for non-American AI.
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.”
The source also notes that OpenAI’s GPT-5.5 Cyber just beat Mythos 5 on certain benchmarks, and that the Trump administration reportedly asked OpenAI to delay the release of GPT-5.6 over security concerns. The reported plan would have the government approve each customer one by one.
That raises the central question for the AI industry: if one powerful model can be pulled into an emergency export control process, others may face the same possibility as their capabilities grow. The current standoff has not answered where the boundary is, who decides it, or how companies should plan around it.
The irony is that the order follows months in which the administration pushed to dismantle AI safeguards and regulation. Yet this has become one of President Trump’s first sweeping regulatory decisions on AI, and cybersecurity leaders cited in the source argue that if regulation is going to happen, this is not the right way to do it.