NSA Director Told Senator: Mythos Breached 'Almost All' Classified Systems in Hours
Summary: The most concrete public account of why the US banned Anthropic's top two models just came from a Senate hearing — NSA Director Rudd confirmed the agency's own red-team found Mythos could compromise classified systems at a speed that alarmed officials.
Key Facts
- Sen. Mark Warner (D-VA), vice chair of the Senate Intelligence Committee, cited NSA Director Gen. Joshua Rudd on June 21
- Rudd told Warner that Mythos breached "almost all" of the NSA's classified systems during a sanctioned red-team test — measured in hours, not weeks
- The test was authorized and cooperative; Anthropic was not accused of hostile activity
- Warner framed the disclosure as reassuring: a model this capable was at least in the hands of a company willing to test before release
- Both Mythos and Fable 5 remain offline globally on day nine of the Commerce Department export directive
Why It Matters
Until now the export ban rested on vague "national security concerns." Rudd's testimony is the first on-record confirmation of the specific capability driving the decision: autonomous penetration of classified government infrastructure at frontier speed. That shifts the policy debate from legal overreach to genuine risk calculus — and signals that US regulators view the most capable AI models as a new class of dual-use asset requiring pre-release government review, regardless of the commercial intent behind them.