Anthropic vs White House: Claude Mythos and Fable 5 blocked

The Trump administration sent an export control directive to Anthropic nearly a week ago, forcing the company to take its most advanced models offline: Claude Mythos and Fable 5. After days of negotiations between Anthropic and the White House, the two remain at odds about how to bring the models back.

What happened?

The export-control order removed access to Anthropic’s most cutting-edge models, locking out all of the company’s customers and many employees, including major clients named in the dispute. All of Anthropic’s customers are locked out, including Apple, Meta, and much of the Fortune 500, and Anthropic says the models had sped up its research and development in recent months. US officials grew concerned earlier this month when they learned Anthropic had shared Mythos with SK Telecom, a South Korean telecom firm alleged to have ties to China, and the White House moved to revoke access immediately after raising the issue.

Negotiations have continued through the week but remain unresolved. The administration has not publicly detailed a specific legal violation by Anthropic; instead, public statements from White House advisers and a post on X outlined the general situation, leaving the dispute defined by its opaqueness.

How are companies reacting and what are they asking for?

Other AI labs and executives say they are preparing to give regulators earlier access and more information to avoid unexpected intervention. Industry leaders told the author that many firms now believe they must provide the White House advance notice and early access to new models. Cohere CEO Aidan Gomez summarized one industry position succinctly: "Advance notice, advance access. I think those are the primary asks that we’ve heard, not just from the US, but others around the world."

The Trump administration’s actions also demanded that Anthropic bar all foreign nationals from accessing Mythos and Fable 5, which the company says prevented many of its own employees from using the models. Amazon CEO Andy Jassy raised a separate concern with US Treasury secretary Scott Bessent that some guardrails on Claude Fable 5 could be circumvented, feeding official worries about the safeguards.

Why does this matter?

The dispute exposes how thin the statutory framework is for governing frontier AI and how much discretion the administration can exercise in practice. President Trump signed an executive order last month that created a voluntary system for labs to submit models for early testing, but the way officials have acted around Anthropic has functionally created an ad hoc licensing posture that companies fear may be treated as compulsory. The result is that leading labs now face a choice: provide early, detailed access to their most advanced models or risk sudden government intervention that can halt commercial and research access overnight.

This dynamic raises real tradeoffs. Requiring early access could reduce surprises for regulators and help target genuine security risks, but it also hands significant power to an administration that, according to one former White House technology official quoted in the reporting, has adopted an "extreme anti-regulatory posture" and has reversed prior federal AI frameworks. The immediate effect is operational: research slowed at Anthropic, customers lost access, and partner collaborations were disrupted.

What to watch

Watch whether other major AI labs start routinely giving the White House early access to new models and whether Washington clarifies the voluntary system created by the executive order into a formal, written process. Also watch whether Anthropic restores access to Mythos and Fable 5 and under what conditions the White House will permit their return. A concrete signal that norms are shifting would be an explicit agency guidance or a written agreement that defines what counts as a national-security risk and what procedural steps companies must take before launch.