Anthropic spent its weekend in a fight it didn’t expect: with the Trump administration, over the AI models it had just spent a week promoting. According to The Verge AI, the company received a US export control directive at 5:21 PM on Friday ordering it to suspend access to its Mythos 5 and Fable 5 models for “any foreign national” inside or outside the US, including Anthropic’s own foreign national employees. The only way to comply, the company determined, was to shut the products down entirely and send executives to Washington to plead its case.
This is one of the most aggressive government interventions into a frontier AI release we’ve seen. The Verge AI reports the administration called Anthropic around 1pm ET Friday with a 90-minute ultimatum: cut off access to Mythos 5 and Fable 5, or face export controls imposed through the Commerce Department.
What set this off
Mythos 5 and Fable 5 are built on the same foundation as Anthropic’s Mythos Preview, a model the company itself called too dangerous to release publicly. Mythos 5 went to a select group of agencies and companies. Fable 5, with extra safeguards, was deemed “safe for general use.”
Then a report suggested those guardrails may have failed. According to The Verge AI, Anthropic believes the government “became aware of a method of bypassing, or ‘jailbreaking’ Fable 5.” Anthropic pushed back hard on how serious it was:
- The company called it a “potential narrow, non-universal” jailbreak, not an existential threat.
- It said the same behavior isn’t unique to Fable 5, claiming the capability is “widely available from other models (including OpenAI’s GPT-5.5).”
- The flaw was reported by an unnamed entity the company declined to name.
Some reports point to Amazon CEO Andy Jassy flagging concerns after Amazon researchers red-teamed Fable 5. That sits at odds with independent red-teamers who said they were impressed by the protections.
The China question
Semafor reported the alarm started over concerns a China-linked group had accessed the technology. But a source in the negotiations told The Verge AI the China rumors traced back weeks, to a large global telecom initially cleared for Mythos Preview access. When the government raised concerns, Anthropic says it revoked that access immediately.
What stands out here is how fast it escalated. Within 15 minutes of the first call, Anthropic executives were talking to the White House. CEO Dario Amodei joined soon after, speaking directly with Treasury Secretary Scott Bessent, Commerce Secretary Howard Lutnick, and National Cyber Director Sean Cairncross, in some cases more than once.
Why this matters
The administration started with a hands-off approach to AI safety. Post-Mythos, The Verge AI reports it’s grown more ambivalent, even as it worries about losing the AI race to China. That tension is now playing out in real time.
The immediate fallout:
- Cybersecurity leaders warn that sidelining Mythos 5 and Fable 5 could hand China an AI advantage, the opposite of the stated goal.
- The move has galvanized international calls for alternatives to American AI systems.
- For Anthropic, the timing is brutal. The company had banked on Mythos to help recover from months of clashes with the Department of Defense.
Anthropic says it pre-briefed the administration on Fable 5, that Commerce ran pre-deployment testing, and that no concerns were raised at the time. An Axios source framed it differently, saying Anthropic “has not done a great job at trying to speak to the administration and appreciate the ideological differences.”
Over the weekend, Anthropic flew in its head of safeguards, its frontier red team lead, and a leading cybersecurity researcher to make its case in person.
The stakes go beyond one company. A government that can force a frontier lab to disable a shipped model on 90 minutes’ notice changes the risk calculus for everyone building at the edge. Watch what the administration decides next, because it could reset how American AI gets released, who gets access, and how much say Washington has over both. More details are available in the original reporting from The Verge AI.