The Trump administration moved Friday to block foreign access to Anthropic's two most advanced artificial intelligence models, Fable 5 and Mythos 5, after a reported security vulnerability raised national security alarms. Commerce Secretary Howard Lutnick delivered the order in a letter to Anthropic CEO Dario Amodei on Friday afternoon, requiring the company to pull both models from any foreign national, whether inside the United States or abroad, effective immediately.

What Triggered the Government's Action?

The Commerce Department acted after a separate company claimed to have successfully "jailbroken" the Mythos model, a technique that bypasses the AI system's built-in safety restrictions. According to an administration official, the government had previously pressed Anthropic to pause the release of the new models, but the company declined, leading to the formal export control letter.

The reported jailbreak works by prompting the model to read a specific codebase and flag software flaws. However, Anthropic disputed the severity of the vulnerability, arguing that the same capability is already available through other publicly accessible models, including OpenAI's GPT-5.5, and is used daily by security researchers and system defenders.

The administration official stated that the models need to stay restricted until the U.S. government's national security apparatus is adequately hardened, a process that could take several weeks. Under the order, a license is required for any export, re-export, or domestic transfer of the models, and Anthropic must apply separately for individually validated licenses. Failure to comply carries potential financial and civil penalties.

How Did Anthropic Respond to the Restrictions?

Anthropic complied with the directive within hours of receiving it at 5:21 p.m. ET on Friday. However, because the company said it has no way to filter users by nationality in real time, it chose to shut both Fable 5 and Mythos 5 down for all customers everywhere, rather than attempting to restrict access only to foreign nationals.

All other Anthropic products, including Claude Opus 4.8, Sonnet, and Haiku, remain online and available to users globally. Anthropic issued a public statement saying it would follow the legal directive but rejected the government's reasoning about the security risk.

"We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people," Anthropic stated in its blog post.

Anthropic, official statement

The company said it has not received evidence that any jailbreak led to actual harm and believes the situation may stem from a misunderstanding. Anthropic indicated it is working toward restoring access to the models.

Understanding the Models at the Center of the Dispute

Anthropic launched Fable 5 and Mythos 5 on June 9, just three days before the government's directive arrived. Fable 5 marked the first time the company had made a Mythos-class model available to the general public. The two models share the same underlying architecture but differ in their restrictions.

• Fable 5 Design: Carries classifiers that block outputs in high-risk categories such as cybersecurity and biology to prevent misuse.
• Mythos 5 Design: Operates with some of those guardrails removed and was limited to a vetted set of organizations through Anthropic's Project Glasswing program.
• Pre-Deployment Testing: Both models were put through thousands of hours of red-teaming by U.S. and U.K. government entities before launch.

Before launch, Anthropic also had a pre-deployment testing partnership with the Commerce Department's Center for AI Standards and Innovation, originally signed in 2024 and renegotiated under the current administration.

How Does This Fit Into Anthropic's Broader Relationship With Washington?

Friday's action adds another layer to an already strained relationship between Anthropic and the federal government. The company is simultaneously listed on a Pentagon blacklist that designates it a supply chain risk, barring it from government use, and now faces a Commerce Department licensing regime that restricts foreign access to its most capable systems.

Earlier this month, President Trump signed an executive order directing pre-deployment testing of frontier AI models. That order is voluntary and does not create a licensing requirement. David Sacks, who served as Trump's White House AI and crypto adviser before stepping down from that role in March 2026 to co-chair the President's Council of Advisors on Science and Technology, had pushed for that voluntary structure, arguing that mandatory licensing would amount to regulatory capture of the largest AI labs.

One administration official told Axios that the president "does not want to hurt the industry and wants innovation to continue," suggesting some tension within the administration over how aggressively to regulate AI development.