The U.S. government ordered Anthropic to disable Claude Fable 5 and Claude Mythos 5 for all users globally on Friday, citing national security concerns. Anthropic complied but publicly disputed the rationale, arguing the decision sets a precedent that could freeze frontier model deployments across the industry.

The directive—received at 5:21 pm ET—was framed as an export control action, but Anthropic’s blog post reveals the actual trigger was a reported jailbreak of Fable 5. According to Anthropic, the government provided only verbal evidence of a “potential narrow, non-universal jailbreak” involving prompting the model to read a specific codebase and identify vulnerabilities. The company maintains this capability is already available in publicly accessible models like OpenAI’s GPT-5.5 and is routinely used by cybersecurity professionals.

Mythos 5, Anthropic’s most capable model, was never publicly released. Instead, it was restricted to roughly 50 vetted organizations—including Amazon, Apple, Google, Microsoft, and CrowdStrike—through Project Glasswing for defensive cybersecurity work. Anthropic claimed Mythos found vulnerabilities in every major OS and browser it tested.

Fable 5, released three days before the shutdown, was Anthropic’s attempt to commercialize Mythos by adding guardrails that block responses in high-risk domains like cybersecurity and biology. The company argued these safeguards operate through independent classifier systems, separate from the model itself, providing defense-in-depth even if a jailbreak bypasses initial refusals.

The irony isn’t lost on industry observers: Anthropic built its brand on safety-consciousness, restricting Mythos precisely because it was too dangerous for public release. That caution now appears to have invited the government scrutiny that threatens its anticipated IPO. As Sam Altman noted in April, when you spend months telling the world your AI is uniquely dangerous, the world tends to listen.

Source: TechCrunch