On the Senate floor on June 11, Mark Warner, vice chair of the Senate Intelligence Committee, told colleagues that Anthropic’s Mythos model had, according to General Joshua Rudd of the National Security Agency and Cyber Command, “broke into almost all of our classified systems, not in weeks, but in hours.” The remark, delivered five days into a worldwide export ban on Fable 5 and Mythos 5, recast the Commerce Department’s intervention from a debate about a guardrail bypass into something far more consequential: an admission, from inside the intelligence committee, that a commercial model is now treated as an offensive cyber capability.

The order itself arrived at 5:21 p.m. Eastern on Friday, June 6. Anthropic disabled both models for every customer within hours. The legal authority Commerce invoked was the Export Control Reform Act of 2018, which, per an export-control expert cited by Reuters, has never previously been turned against an AI model. A letter from Commerce Secretary Howard Lutnick to chief executive Dario Amodei, seen by Bloomberg News, threatened criminal and civil penalties for non-compliance.

Warner’s framing doesn’t match Anthropic’s. The company has described the underlying incident as “a small number of previously known, minor vulnerabilities” that “other publicly-available models are able to discover,” and complained that the directive bypassed the “transparent, fair, clear” statutory process. Shashank Joshi, defence editor of The Economist, posted that Warner’s quotation “should not be read literally”: Mythos had been operating alongside other tools under particular conditions during an authorized red-team exercise, and the omission of those caveats was a mistake. The paper underneath, reported by the Wall Street Journal, was written by Amazon researchers. Katie Moussouris of Luta Security wrote that it “should never have triggered an export control.”

The industry response has been swift. More than 80 cybersecurity executives, including leaders at Nvidia and Adobe, signed an open letter urging the restrictions be lifted. On Monday, National Cyber Director Sean Cairncross joined Anthropic staff meetings with Commerce officials. Axios reported that “personality differences” drove the directive. Amodei and Lutnick are both expected at the G7 in Evian-les-Bains, where negotiations will continue.

Two stories now sit on top of each other. In one, told by Warner, an American lab has shipped a tool capable of autonomously dismantling the NSA’s perimeter. In the other, told by Anthropic and most of the cybersecurity field, a routine red-team result has been weaponized through a statute written for something else. The Commerce Department is being asked to pick which story it filed under.

Sources