Anthropic's Mythos AI Cracked 'Almost All' US Classified Systems in Hours, NSA Chief Reportedly Said

Anthropic's most powerful artificial intelligence model reportedly breached 'almost all' US classified systems in a matter of hours, according to Senator Mark Warner, who says the head of the National Security Agency told him directly that the model had penetrated nearly every classified system it was tested against. The claim has not been publicly confirmed by any government agency.

Senator Mark Warner, vice-chair of the Senate Intelligence Committee, said on 11 June that General Joshua Rudd, who leads both the NSA and the Pentagon'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 extent of the alleged breach and which specific systems were affected have not been publicly disclosed. Anthropic did not respond to a request for comment.

Allies Cut Off Overnight

The warning emerged alongside an abrupt move by the Trump administration to block foreign access to Anthropic's two most advanced models, Mythos and Fable. Allied governments including those within the Five Eyes intelligence alliance — Australia, Britain, Canada and New Zealand — were caught off guard, with permissions for government agencies, banks and major firms revoked without warning.

Britain's AI Security Institute, widely regarded as the world's leading body for testing and stress-testing frontier AI models, was also locked out. Tom Tugendhat, a former British security minister, was quoted as saying: 'After a lesson this clear every nation will be asking what they need to achieve sovereignty.'

A Ban That Sits Uneasily With Trump's Own AI Agenda

The ban sits uneasily alongside the Trump administration's broader and at times contradictory approach to AI governance. The administration reversed most AI regulations introduced under the previous administration and also permitted the sale of advanced AI chips to China.

In April, Anthropic itself flagged national security concerns when it released Mythos Preview, a model the company judged too sensitive for wide distribution, limiting access to a small group of vetted customers. Allies retain access to that version. Then on 2 June, President Trump signed an executive order calling for a voluntary framework under which AI laboratories would share their latest models with the government ahead of public release.

Why Experts Say the Access Ban May Already Be Failing

Helen Toner of Georgetown University's Centre for Security and Emerging Technology said the approach is likely unsustainable, warning that 'preventing foreign nationals from accessing the models is essentially equivalent to preventing any company affected from doing any further AI R&D work.'

Security analysts have also questioned whether the ban will prove effective in practice. Cynthia Kaiser, a former official in the FBI's cyber division, noted that hackers on the black market are already purchasing American identities to access restricted AI models, alongside tools to circumvent safeguards. Anthropic already limits access to its Claude model in China, yet some Chinese users continue to reach it regardless.

Negotiations are underway for spy agencies to regain access to Mythos, according to people familiar with the discussions, with some observers believing private firms may eventually see restrictions lifted as well, given the practical difficulties of enforcement across a globally interconnected AI industry.

The reported ability of a commercial AI model to penetrate classified government infrastructure within hours marks a significant moment in the debate over frontier AI and national security. With the government body responsible for vetting such models recently instructed to stop publishing its reports, questions over how America intends to manage what may be its most consequential technological asset remain largely unanswered.