Export Controls Cut NSA Off From Anthropic's Mythos 5 AI

Parts of the NSA lost access to Anthropic’s Mythos 5 model after the Trump administration imposed export controls on the company this month, citing national security concerns. Analysts were notified on Friday that access was being pulled. The trigger was a government directive ordering Anthropic to suspend Mythos 5 and Fable 5 for any foreign national — including noncitizen employees working inside the United States. Anthropic said it could not enforce that restriction selectively, so it disabled the models for all customers rather than carve out specific users.

The affected access ran through Project Glasswing, Anthropic’s restricted early-access program for Mythos aimed at security researchers, companies, and government-linked defenders. Launched in April, the program had grown to roughly 150 organizations across more than 15 countries before the shutdown. The NSA may still be able to lean on earlier model versions under prior arrangements, but with company support, updates, and modifications now curtailed.

Much of the public attention came from a June 11 hearing where Sen. Mark Warner relayed that NSA Director Gen. Joshua Rudd had described Mythos breaking into ‘almost all’ classified systems ‘not in weeks, but in hours.’ Officials later clarified to reporters that this referred to a sanctioned red-team exercise designed to test the model’s offensive capability — not an actual intrusion into NSA networks. The episode underscores the tension between deploying frontier AI for defensive work and the export-control regime now constraining who can touch the most capable models.