Anthropic's Mythos AI breaches NSA systems in hours

Anthropic's most powerful AI model, Claude Mythos, was reported to have penetrated classified networks of the U.S. National Security Agency within hours. NSA director Joshua Rudd confirmed the intrusion, and Senate Intelligence Committee vice‑chair Mark Warner said the breach was disclosed on 11 June.

Following the incident, the U.S. administration ordered Anthropic to suspend foreign access to its Mythos and Fable models and the White House demanded the immediate removal of the most advanced versions, citing national‑security risks. Anthropic had placed several engineers inside the agency to guide the technology’s use and had previously assisted the NSA in deploying Mythos for cyber‑offensive operations. Access to the model is now limited to a small partner group called Glasswing, and the agency no longer has direct access.

Officials warned that the capability could be leveraged against nations such as China and Iran and highlighted concerns over AI “jailbreak” and recursive self‑improvement, prompting tighter oversight of advanced AI systems.