Anthropic’s Mythos artificial intelligence (AI) model reportedly gained access to nearly all classified systems operated by the National Security Agency (NSA) within hours during an authorized security evaluation, according to a U.S. senator.
The claims emerged following a Senate Intelligence Committee discussion in which Sen. Mark Warner, vice chairman of the panel, said Gen. Joshua Rudd, who leads both the NSA and U.S. Cyber Command, described the results of a June 11 red-team exercise involving Anthropic’s Mythos model.
“On June 11th Mark Warner, the vice-chair of the Senate Intelligence Committee, said that General Joshua Rudd, who leads the National Security Agency and the Pentagon’s Cyber Command, had told him that Mythos ‘broke into almost all of our classified systems, not in weeks, but in hours’.” reported The Economist.
The quote went viral about a week later across multiple social media platforms, fueling claims that Anthropic’s model had “hacked the NSA.”
In response, the author issued a public statement clarifying that the narrative was false, adding that the activity took place during an authorized internal security test in which Mythos was paired with defensive cybersecurity tools under controlled conditions. Warner’s claims have not been confirmed by the NSA, U.S. Cyber Command or other government agencies.
An update. A US official tells me that Sen. Warner misunderstood the NSA director Gen. Rudd in this case. Rudd did use the ‘hours, not weeks’ wording, but the use of Mythos in this context was—as widely assumed—part of a red-teaming effort, i.e. testing the security of internal… https://t.co/DiPITXmo0Q
— Shashank Joshi (@shashj) June 22, 2026
Cyber security concerns surrounding Anthropic’s newest AI models, Fable 5 and Mythos 5, have emerged since their launch earlier this month.
On June 12, a U.S. Department of Commerce directive required Anthropic to limit access to its Fable 5 and Mythos 5 models to U.S. citizens, citing national security concerns. The decision followed reports that researchers had successfully jailbroken Anthropic’s Claude Fable 5 model using techniques that bypassed its built-in safety safeguards and enabled the generation of outputs that would normally have been blocked.
According to Semafor, the decision was also partly prompted by Amazon CEO Andy Jassy, who reportedly raised concerns with White House officials about security risks and indications that a China-linked group may have gained access to Mythos.
Anthropic responded to the directive by suspending access to both models worldwide, saying it could not effectively enforce nationality-based restrictions on individual users.
Despite complying with the order, Anthropic disagreed with the decision to withdraw the AI models. In a statement, the company said the reported jailbreak did not warrant recalling the systems and argued that similar vulnerabilities exist in competing AI systems, including OpenAI’s GPT-5.5.
“We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people. If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers,” the company said.
“As we have stated publicly, we believe the government should have the ability to block unsafe deployments, as part of a statutory process that is transparent, fair, clear, and grounded in technical facts. This action does not adhere to those principles,” Anthropic added.
