Ten days after a Commerce Department letter arrived at Anthropic at 5:21 p.m. Eastern on June 12, the company’s two most capable models, Claude Fable 5 and Claude Mythos 5, remain offline for every customer in the world. The directive cited national security authorities and barred distribution to foreign nationals. Because that category includes non-citizen Anthropic employees inside the United States, the company concluded it had no operational path short of a global suspension. Opus 4.8 and earlier Claude releases continue to run.

Fable 5 had been public for roughly three days when the letter landed. Vals AI benchmarks rated it the most capable publicly available model at release. Mythos, previewed in April, had been distributed under a restricted program called Project Glasswing, first to about 50 vetted organizations, then expanded to roughly 150 across 15 countries. It’s the public version, the one Anthropic says reaches hundreds of millions of people, that the government has effectively recalled.

Anthropic’s public statement disputes the basis for the order in unusually direct terms. The company says the Commerce Department offered only verbal evidence of a narrow jailbreak that “essentially consists of asking the model to read a specific codebase and fix any software flaws,” a capability Anthropic says is “widely available from other models, including OpenAI’s GPT-5.5.” Applying the same standard across the industry, the company argued, “would essentially halt all new model deployments for all frontier model providers.” Thousands of hours of red-teaming preceded launch. “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,” the statement reads.

The defender community has aligned with that view. On June 15, dozens of cybersecurity experts published an open letter calling the order “dangerous” and accusing the government of having “taken the best models away from defenders.” Katie Moussouris, among the signatories, said the cited behavior “is not a guardrail bypass” but the ordinary “find, fix, and test loop defenders run every day.”

The order doesn’t arrive in a vacuum. In February, President Trump directed federal agencies to stop using Anthropic’s models. In early March, the Pentagon designated the company a “supply chain risk,” a label Anthropic is contesting in federal court after refusing contract language permitting use “for any lawful purpose” and seeking carve-outs for autonomous weapons and mass domestic surveillance. Earlier in June, the company confidentially filed for an IPO reportedly valued at $965 billion.

A safety-forward lab built a regulatory posture around restraint, and the state took it at its word.

Sources