NSA director says AI model 'Mythos' breached classified systems in hours
Tags AI · Security · Infrastructure
The NSA director stated publicly that Anthropic's Claude Mythos AI model was able to penetrate "almost all" of the agency's classified systems within hours during testing, according to a report in The Economist. The comment provides the most direct government assessment yet of the cybersecurity capabilities of advanced AI models and helps explain the Trump administration's decision to block Anthropic from distributing Mythos and Fable 5. The assessment has significant implications for how governments evaluate AI model capabilities and the emerging intersection of AI safety and national security.
Technical significance
The NSA director's statement is the most authoritative government confirmation that frontier AI models possess offensive cybersecurity capabilities at a level that threatens classified networks. This validates the concerns driving export control debates and suggests that AI model evaluations will increasingly need to assess dual-use cyber capabilities as a standard benchmark, with profound implications for how AI labs release and govern their most powerful models.