Anthropic's Mythos model found vulnerabilities in classified US government systems, official says
A U.S. official confirmed that Anthropic's Mythos AI model identified vulnerabilities in highly sensitive and secure U.S. government computer systems during a testing exercise.
The Atomic take
If an AI model can probe classified systems and surface real weaknesses, the same capability becomes a dual-use weapon the moment it lands in adversarial hands, which raises the stakes on how Anthropic gates access to Mythos. The fact that a US official is willing to confirm this on the record suggests Washington is moving from quiet testing toward openly leaning on commercial models for offensive and defensive security work. Watch whether this leads to a formal procurement contract or new export and access controls, and whether other labs disclose similar government red-team results.
Read the full story at CNBC →Know where humanity is headed.
Five forces. One brief. Free, every Monday morning.
Get The Atomic Five