Claude Fable 5 and Mythos 5 entered a fourth day offline on Tuesday, with no public movement after Monday’s closed-door session between Anthropic engineers and Commerce Department officials in Washington. According to NBC News, it’s the first time the federal government has forced a leading AI company to pull its systems from public use.
The directive landed Friday afternoon. At 1:00 p.m. ET on June 12, the government called Anthropic and instructed it to disable both models; a formal letter followed around 5:30 p.m. Fable 5 had been live for three days, released June 9 and described by Vals AI’s benchmarks as the most capable public model on the market at launch. Mythos 5, previewed in April, had been kept inside Project Glasswing, a roughly 50-organization program that includes Amazon, Apple, Google, Microsoft and CrowdStrike.
The order’s origin story is unusually legible for an export-control action. NBC News reports the move traces at least in part to a passing reference, on a call between Amazon CEO Andy Jassy and Trump administration officials, to a security finding about Fable 5. A White House official told CNBC that Jassy was among several tech leaders who raised concerns. An Amazon spokesperson called it “not uncommon for governments to seek out counsel on potential security risks.” Amazon has already invested $8 billion in Anthropic and committed up to $25 billion more, which makes the Jassy-as-trigger detail load-bearing in any read of elite incentives here.
Anthropic, in a statement on its site, says the government’s evidence amounts to a “potential narrow, non-universal jailbreak” that prompts Fable 5 to “read a specific codebase and fix any software flaws,” a capability the company says is widely available and routinely used by defensive cybersecurity teams. The order’s scope, per Anthropic, reaches “any foreign national, whether inside or outside the United States.” So far the government has offered only verbal evidence.
The company’s framing is pointed. “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 wrote, adding that the standard, if applied industry-wide, “would essentially halt all new model deployments for all frontier model providers.” TechCrunch noted Anthropic’s specific comparison to OpenAI’s GPT-5.5.
Monday’s meeting, characterized by a source close to the company as an attempt to resolve the dispute, produced no announcement. Neither Bloomberg nor CNBC identified which officials attended. Anthropic’s separate litigation over a Defense Department supply-chain designation, per NBC, remains ongoing.
The structural read is straightforward. A safety-forward lab built its public identity around disclosing model risks, and the disclosure regime it championed has now been used, via an investor’s phone call, to take its flagship offline. The precedent is the story.
Sources
- https://www.nbcnews.com/tech/security/anthropic-fable-5-ai-offline-trump-order-administration-claude-rcna350117
- https://www.bloomberg.com/news/articles/2026-06-15/anthropic-set-to-meet-trump-officials-over-ai-security-concerns
- https://www.cnbc.com/2026/06/15/anthropic-mythos-trump-ai.html
- 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.anthropic.com/news/fable-mythos-access