Anthropic's Mythos AI broke into almost all NSA classified systems in hours
Pierluigi Paganini reports the Senate-testimony bombshell: per Senator Mark Warner, General Joshua Rudd told lawmakers that Anthropic's unreleased Mythos model penetrated nearly all NSA and US Cyber Command classified systems 'not in weeks, but in hours'. The article is careful to flag everything as unverified claims relayed through testimony, but the downstream facts are concrete enough — the administration ordered Fable 5 and Mythos 5 restricted to US citizens, both models went dark globally overnight for want of any workable nationality check, and Five Eyes partners mid-evaluation were locked out along with the UK AI Security Institute. Whichever way you read it — capability disclosure or policy improvisation — it is the clearest public signal yet of where frontier-model offensive capability sits, and of how little process exists for handling it.
Was this useful?