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

Anthropic’s flagship Mythos AI model reportedly infiltrated nearly all of the National Security Agency (NSA) ‘s classified systems within a few hours during an authorized red-team evaluation on June 11. This incident now seems to be the main reason for a broad U.S. government directive on export controls issued the following day.

Senator Mark Warner, Vice Chairman of the Senate Intelligence Committee, disclosed that General Joshua Rudd, who simultaneously leads the NSA and U.S. Cyber Command, told him directly that Anthropic’s Mythos model “broke into almost all of our classified systems, not in weeks, but in hours.”

The statement, first reported by The Economist, has not been formally confirmed by any government agency, but has rapidly reshaped the narrative around Washington’s decision to pull Anthropic’s two most advanced models from public access.

The disclosure recontextualizes the June 12 Commerce Department directive, which barred every foreign national, including non-citizen employees within Anthropic itself, from accessing Fable 5 and Mythos 5.

Anthropic subsequently suspended both models for all customers. Crucially, this marks the first time the United States has applied export controls directly to an AI model rather than to the hardware or chips powering it, a landmark regulatory precedent in AI national security governance.

Allied governments within the Five Eyes intelligence alliance, including Australia, Britain, Canada, and New Zealand, were reportedly caught off guard, with permissions for government agencies, banks, and major firms revoked without prior warning.

Anthropic disputes the government’s stated rationale. The company contends that the cited trigger was a narrow jailbreak one that other leading models, including GPT-5.5, also exhibit and characterizes the wholesale recall as a disproportionate overreaction.

In Anthropic’s account, the flagged behavior amounted to asking the model to analyze a codebase and fix identified issues, not a genuine autonomous offensive intrusion. The company is actively working to restore access and is preparing a collaborative risk management framework with the White House.

Follow us on Google News, LinkedIn, and X to Get More Instant Updates.