NSA Loses Anthropic Mythos Access After June Export-Control Order

The National Security Agency lost access to Anthropic's Mythos 5 after the Trump administration's June export-control directive forced the company to pull back its most advanced models, The New York Times reported Tuesday, citing U.S. officials. The cutoff hit analysts who had been testing the cybersecurity model inside the agency, where officials said it had found software weaknesses faster than expected.

The new reporting narrows a story that had moved through Washington in a more alarming form. Sen. Mark Warner, the top Democrat on the Senate Intelligence Committee, said at a June 11 hearing that NSA Director Gen. Joshua Rudd had told him Mythos "broke into almost all of our classified systems, not in weeks, but in hours." Officials later told the Times and the Associated Press that the statement referred to a controlled red-team exercise, not an outside breach of NSA networks.

Analysts were notified after the directive

Parts of the NSA lost Mythos access after the administration acted against Anthropic this month, Nextgov/FCW reported, citing two people familiar with the matter. Some analysts were told Friday that access would end, one of the people said. The agency may still be able to use earlier versions under prior arrangements, although support, updates or modifications could be more limited, the second person told the outlet.

Anthropic said on June 13 that it had received a government directive at 5:21 p.m. Eastern requiring it to suspend Fable 5 and Mythos 5 access for any foreign national, including noncitizen employees inside the United States. The company said it disabled the models for all customers because it could not comply selectively. The order came days after the company released Fable 5 more widely while keeping Mythos tightly restricted.

Warner's quote came from a red-team test

The red-team detail matters because it changes the technical claim. A U.S. official told the Associated Press that Anthropic's model identified vulnerabilities in highly sensitive government systems within hours during a testing exercise, but that did not mean the model exploited them within that time. The Times reported that NSA red teams began inside classified systems designed to be reachable only from certain computers and cut off from the wider internet.

That setup made the test useful for defenders and poor evidence of a model breaking into classified systems from scratch. Red teams are authorized testers who probe networks so an agency can fix problems before real attackers use them. Officials still told the Times that NSA analysts were struck by Mythos's performance in the controlled environment, where it exceeded expectations for finding flaws.

Glasswing becomes the access problem

The access loss appears tied to Project Glasswing, Anthropic's restricted program for early Mythos use by security researchers, companies and government-linked defenders. Anthropic launched the effort in April, later widening it to roughly 150 organizations in more than 15 countries. Nextgov cited an Economist editor's post saying NSA red teams lost access because their authority to use Mythos came through Glasswing.

The cutoff follows months of conflicting U.S. government positions on Anthropic. The Pentagon labeled the company a supply-chain risk earlier this year after a fight over military-use terms, while the NSA and other agencies kept testing the company's most cyber-capable model. The Implicator previously covered that split when the agency was reported to be using Mythos despite the Pentagon label and again when Anthropic embedded engineers at the NSA.

Five Eyes agencies warn on timing

The timing sharpened the contrast with cyber agencies' warning that defenders need to adopt AI quickly. On Monday, the United States, Britain, Canada, Australia and New Zealand issued a Five Eyes statement warning that frontier AI models could change offensive and defensive cyber work on a timeline measured in months. "Adversaries are already using AI to move faster and more effectively," the agencies said. "Defenders must do the same."

Anthropic and government officials are still trying to resolve the export-control issue, according to the Times. A classified contract that would let the NSA use Anthropic technology for intelligence analysis and vulnerability detection has not been finalized, and some Pentagon officials want the agency to work with other models. For now, the practical result is narrower: the same agency that tested Mythos against its own systems has lost at least part of the tool it was using to find what needed fixing.

US Export Controls Force Anthropic to Pull Fable 5 and Mythos 5 Days After Launch

Commerce Secretary Howard Lutnick sent Anthropic chief executive Dario Amodei a letter on Friday placing the company's Fable 5 and Mythos 5 models under export controls, barring their use by any forei

The Implicator

Anthropic Embeds Engineers in the NSA to Deploy Mythos for Offensive Cyber

Anthropic put about six engineers inside the NSA to deploy Mythos, the cyber model it calls too dangerous to release, for offensive operations, the FT reported.

The Implicator

NSA Tests Anthropic Mythos on Microsoft Software, Report Says

Bloomberg says the NSA is testing Claude Mythos Preview against Microsoft software while the White House weighs wider federal access.

The Implicator

Marcus Schuler

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai