U.S. government bans Anthropic's most powerful AI security models, prompting 76 cybersecurity experts to protest that the restriction harms defenders more than adversaries.

What happened

The U.S. government ordered Anthropic on Friday to limit exports of its Fable and Mythos models citing national security concerns. In response, Anthropic suspended access to both models worldwide. Seventy-six cybersecurity experts, including former Facebook security chief Alex Stamos and cryptographer Jon Callas, published an open letter asking the government to lift the order, arguing the ban removes the best tools from security researchers who use AI to find and fix software vulnerabilities.

Why it matters

The government's action appears to have been triggered by a report showing a method to bypass Fable's safeguards, though cybersecurity expert Katie Moussouris—who reviewed the paper—argues the method is not a genuine jailbreak but rather normal defensive work (asking the model to fix code with vulnerabilities). Moussouris and the letter signers contend that weakening these models for defense purposes is counterproductive, since the ability to find, fix, and test bugs is exactly what defenders need daily.

What to watch

The letter argues the demonstrated method can be replicated on OpenAI's GPT-5.5, Anthropic's publicly available Claude Opus 4.8 and Sonnet, and Chinese models like Kimi 2.7, raising questions about whether the ban's rationale is consistent or if export controls will expand to other models. The experts also call for transparent, science-based regulation created through a democratic rule-making process.