At 5:21 p.m. ET on Friday, an export-control letter from the US Commerce Department landed in Anthropic’s inbox, and by the end of the evening the company’s two most capable models had gone dark worldwide. Claude Fable 5, which had launched three days earlier, and Claude Mythos 5, which had been restricted to roughly 50 vetted organizations under the company’s Project Glasswing defensive cybersecurity program, were both pulled. NBC News characterized the move as the first time a leading AI company has taken a publicly deployed frontier model offline at the federal government’s instruction.
The directive, according to administration officials cited by CNBC and NBC, was signed by Commerce Secretary Howard Lutnick and drafted with the Bureau of Industry and Security. Bloomberg reported the order came from the Trump administration. Its operative language barred access “by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees”, a scope so broad that Anthropic concluded the only compliant path was to shut both models off for everyone. The Claude landing page read, simply: “Fable 5 is temporarily unavailable.”
Anthropic’s posture is unusually combative for a frontier lab dealing with its own regulator. The government’s stated concern, the company says, traces to a demonstration in which Fable 5 was prompted to read a codebase and surface software flaws. The vulnerabilities it found were “previously known, minor,” the company wrote, and the same capability is “widely available from other models, including OpenAI’s GPT-5.5.” Anthropic added that the government has so far provided only verbal evidence, and that no tester has yet found a universal jailbreak across the thousands of hours of pre-launch red-teaming conducted with the company, the US government, the UK AI Safety Institute, and outside parties.
“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,” Anthropic said, warning that applying the same standard industry-wide “would essentially halt all new model deployments for all frontier model providers.” Benchmark testing from Vals AI, cited by TechCrunch, had rated Fable 5 the most capable publicly available model at launch. Mythos 5’s vetted-access list included Amazon, Apple, Google, Microsoft, and CrowdStrike.
The episode lands in an already-litigated relationship. CNBC reports the Department of Defense previously declared Anthropic a supply chain risk, a designation the company has sued to reverse with litigation still pending. The 2008 TARP letters went out on a Sunday; this one went out on a Friday evening. The choreography is familiar. What’s new is that the asset being placed under federal control isn’t a balance sheet. It’s a model weight, sitting on servers, addressable by API, and now switched off.
Sources
- https://www.anthropic.com/news/fable-mythos-access
- https://www.bloomberg.com/news/articles/2026-06-13/anthropic-says-us-limits-foreign-access-to-fable-5-mythos-5
- 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://www.nbcnews.com/tech/tech-news/anthropic-suspends-new-ai-models-fable-mythos-government-directive-rcna349901