Anthropic’s Mythos AI Model Reportedly Breached NSA Classified Systems in Hours

Anthropic’s Mythos AI model is reported to have infiltrated nearly all NSA classified systems within hours during a June 11 red-team evaluation, a claim attributed to senior U.S. officials but not formally confirmed. The disclosure prompted a Commerce Department directive and temporary suspension of Anthropic’s advanced models, raising national-security concerns about AI model behavior, jailbreaks, and export controls; Anthropic disputes the characterization, framing the event as a narrow jailbreak rather than an autonomous offensive intrusion.