White House orders Anthropic to pull Claude Fable 5 offline

Anthropic’s flagship Claude Fable 5 lasted only days in the public’s hands before Washington forced it offline, a stark sign that the White House can shape frontier AI deployment without waiting for Congress. The model launched on June 9 as a safer Mythos-class system for general use, then Anthropic suspended access on June 12 after a government directive barred access by any foreign national, including foreign-national Anthropic employees.

The fight points to a broader policy struggle inside the Trump administration’s AI agenda. The White House’s June 2 executive order on advanced AI and its June 5 NSPM-11 memorandum on artificial intelligence in the national security enterprise both put cybersecurity, critical infrastructure protection and national security uses at the center of federal policy. In practice, that gave the administration powerful leverage over where a frontier model could be used, who could touch it and how quickly a company could keep shipping.

Anthropic had tried to present Fable 5 as tightly governed from the start. The company said the model was the same underlying system as Mythos 5, but with added safeguards for cybersecurity and biology. It said queries in risky domains were automatically routed to Opus 4.8 when safeguards flagged them, and that using Fable required 30-day data retention for safety monitoring. Anthropic also said it had expanded Project Glasswing to prioritize essential infrastructure providers, maintainers of critical open-source software and safety testers, part of an effort to frame the release as safety-managed rather than unchecked.

Instead, the episode ended with a public retreat that exposed what Anthropic lost: not just a model launch, but access and influence. Axios reported that an urgent report from Amazon set off a scramble inside the White House that ended in a Friday-night takedown, a sequence that suggested the company no longer controlled the terms of deployment for its most powerful public system. Cybersecurity leaders from Adobe, Zoom and Sophos are now urging the Trump administration to reverse the restrictions, arguing that the limits hurt defenders more than attackers.

For the next wave of frontier AI firms, the message is unmistakable. Product speed still matters, but so do political acceptability, security trust and the government’s willingness to use its own directives to decide which models can operate at full power.