White House shuts down Anthropic's Mythos, Fable models

What happened

On Friday evening, the White House placed export control restrictions on Anthropic's Mythos 5 and Fable 5 models, preventing foreign governments and nationals from using them and forcing Anthropic to completely shut off access to both products. Several tech executives, including Amazon president and CEO Andy Jassy, had raised concerns within days of launch that the models could be jailbroken—meaning their safety guardrails could be bypassed—posing a cybersecurity threat.

Why it matters

Even after several days of statements and negotiations, the actual basis for the action remains murky. One account claims Amazon found a jailbreak method specific to Fable, while others say the same results could be achieved with OpenAI's ChatGPT 5.5, or that a China-linked group may have accessed Mythos (though no confirmed jailbreak occurred). Some observers suggest the decision reflects political friction—that Anthropic and its CEO Dario Amodei have not aligned with the Trump administration's expectations and messaging, creating a 'vibes' conflict independent of the security claims.

What to watch

Former White House AI adviser Dean Ball noted that frontier AI companies now operate under a system where government approval is effectively required to operate, regardless of what law or prior statements say—and Anthropic currently has no faction within the administration rushing to defend it. How the company recovers and whether other frontier AI firms adapt their approach to the administration's expectations will signal how political alignment shapes AI regulation going forward.