Anthropic's advanced AI model, Mythos, identified security flaws in highly classified US government systems within hours during a testing exercise conducted with intelligence agencies under the company's Project Glasswing initiative.
Senator Mark Warner disclosed the results at a recent hearing, citing NSA and Cyber Command chief General Joshua Rudd, who confirmed the AI "broke into almost all of our classified systems, not in weeks but in hours."
The disclosure comes as the Trump administration issued a directive compelling Anthropic to block foreign nationals from accessing its latest Fable 5 and Mythos 5 models, intended as a cybersecurity safeguard. Anthropic complied but argued the restrictions were unwarranted.
The move drew opposition from over 100 cybersecurity experts and companies including Adobe and Nvidia, who warned that removing advanced AI defense capabilities could advantage US adversaries.