At 5:21 p.m. ET on Friday, Anthropic received a U.S. export control directive ordering it to disable global access to Fable 5 and Mythos 5, the first time Washington has reached into a frontier lab’s serving stack and pulled a deployed model off the internet for every customer at once. By June 12 the cutoff was live, including for AWS customers running the models in production, an outage that AWS itself acknowledged in a status update flagged by TechCrunch.
The model on the table had been public for three days. Fable 5, the guardrailed variant of Mythos 5, launched earlier that week and was rated by Vals AI as the most capable publicly available model at launch, ahead of OpenAI’s GPT-5.5. Mythos 5, previewed in April, had never gone wide. CNBC and TechCrunch reported it was restricted to roughly 50 vetted organizations, including Apple, Google, Microsoft, CrowdStrike, and the Department of Defense, under an Anthropic program called Project Glasswing.
The how is where the story gets uncomfortable for the lab whose entire brand is safety. According to The Wall Street Journal, with similar accounts from The Information and Reuters, Amazon CEO Andy Jassy told Treasury Secretary Scott Bessent and other officials that Amazon’s own researchers had used Fable 5 to extract information usable in cyberattacks. Amazon confirmed the outreach in spirit, telling reporters it’s “not uncommon for governments to seek our counsel on potential security risks.” David Sacks, co-chair of the President’s Council of Advisors on Science and Technology, posted on X that “a highly credible trusted partner of both Anthropic and the USG” surfaced the jailbreak and that the administration had asked CEO Dario Amodei to fix the flaw or de-deploy Fable 5, which Sacks said Amodei refused. Semafor, cited by Al Jazeera, added that the order was issued partly over suspicions a China-linked group had accessed the model.
Anthropic’s public statement doesn’t concede the premise. The company says the government produced “verbal evidence of a potential narrow, non-universal jailbreak” and that internal review found only “a small number of previously known, minor vulnerabilities.” It then draws the structural line: “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,” warning the precedent “would essentially halt all new model deployments for all frontier model providers.”
That second sentence is the real news. For three years the frontier labs have argued that voluntary safety disclosures and red-team partnerships with government were the alternative to hard regulation. The recall inverts the trade. Anthropic’s willingness to publish capability evaluations, combined with a hyperscaler partner walking a finding directly to Treasury, gave the executive branch both a pretext and a delivery channel. The safety-pilled lab built the legibility that made it the first to get switched off.
Sources
- https://www.anthropic.com/news/fable-mythos-access
- https://techcrunch.com/2026/06/12/anthropics-safety-warnings-may-have-just-backfired-the-government-has-pulled-the-plug-on-its-most-powerful-ai/
- https://www.cnbc.com/2026/06/12/anthropic-disables-access-to-fable-5-and-mythos-5-to-comply-with-government-directive.html
- https://techcrunch.com/2026/06/13/amazon-ceo-reportedly-raised-anthropic-model-concerns-before-government-crackdown/
- https://www.aljazeera.com/amp/news/2026/6/14/us-asks-anthropic-to-block-global-access-to-top-ai-models-why-it-matters