Mythos broke into almost all of NSA's classified systems in hours

A senator says Anthropic's Mythos penetrated almost all of the NSA's classified systems during authorized testing. Days later, Washington forced the company to disable its most capable models for everyone.

A week after the US government forced Anthropic to switch off its most powerful AI models, a possible explanation has surfaced, and it is striking. According to The Economist, Senator Mark Warner, vice-chair of the Senate Intelligence Committee, told a hearing that General Joshua Rudd, who leads both the National Security Agency and US Cyber Command, said the company's Mythos model "broke into almost all of our classified systems, not in weeks, but in hours."

One caveat matters up front. By the accounts circulating, that result came during authorized red-team testing, a sanctioned exercise to probe the model's offensive cyber capability, not a hostile intrusion. But the demonstration helps explain why the government has started treating the model less like software and more like a weapon.

On June 12, the Commerce Department issued an export-control directive ordering Anthropic to suspend all access to Fable 5 and Mythos 5 for any foreign national, whether inside or outside the United States, including the company's own non-citizen employees. According to the Wall Street Journal, Commerce Secretary Howard Lutnick sent the letter to Anthropic chief executive Dario Amodei. Faced with a rule it could not apply selectively, Anthropic disabled both models for every customer worldwide, while leaving its other models, including Claude Opus 4.8, untouched. The company said it received the directive at 5:21pm Eastern and that the letter gave no detailed explanation of the security concern.

It is, by most accounts, the first time the United States has export-controlled an AI model itself rather than the chips used to train one.

Anthropic disputes the rationale. In its statement, the company said its understanding is that the trigger was a method of "jailbreaking" Fable 5, the safeguarded public version of the Mythos model. Anthropic described the technique as narrow and non-universal, said it amounted to little more than asking the model to read a codebase and fix flaws, and argued that other publicly available models, including OpenAI's GPT-5.5, can do the same. Recalling a commercial product used by hundreds of millions of people over that, the company said, would halt new model releases across the industry if applied as a general standard.