NSA Loses Mythos 5 Red-Team Access After Anthropic Export Controls
TL;DR
- Parts of the NSA lost access to Anthropic's Mythos 5 on Friday, cutting off red teams whose authority came through Project Glasswing, launched in April.
- Secretary of Commerce Howard Lutnick's directive banned foreign national access to Fable 5 and Mythos 5, including foreign-born Anthropic employees.
- Anthropic characterized the triggering jailbreak as 'narrow' and 'non-universal,' and disputed that Mythos 5's capabilities exceeded those in other publicly available models.
The NSA analysts running authorized red-team exercises against their own classified networks had been using Anthropic's Mythos 5 to do it. According to The New York Times and reporting from Nextgov/FCW, some of those analysts were notified on Friday that access was gone, after Secretary of Commerce Howard Lutnick directed Anthropic to suspend foreign access to Fable 5 and Mythos 5.
The model had earned its place in that workflow. Sen. Mark Warner described NSA Director Gen. Joshua Rudd telling him that Mythos broke into "almost all of our classified systems, not in weeks, but in hours," during authorized testing. The NSA was using the model as a tool to probe its own networks for weaknesses, not as an adversary. That red-team access was structured through Project Glasswing, an effort launched in April that Nextgov/FCW reports had expanded to roughly 150 organizations in more than 15 countries. Losing it arrives as the Five Eyes alliance warns that frontier AI models could sharply change the cyber threat landscape within months, not years, by helping attackers and defenders move faster.
The honest caveat is that the ban's triggering event may have been narrower than the resulting access loss. CyberScoop reported that Anthropic understood the government to have identified a "narrow, non-universal jailbreak" involving prompting the model to analyze code for flaws. Anthropic disputed the severity, arguing those capabilities were already available in other publicly accessible models, including OpenAI's GPT-5.5. The export controls apply to foreign nationals anywhere in the world, including foreign-born Anthropic employees, a scope wider than the jailbreak itself.
What the reporting does not resolve is whether the NSA's limited retained access to earlier Mythos versions can substitute for Mythos 5 in practice. Anthropic's proposed identity verification policy, set to take effect July 8, could restore domestic access by allowing the company to verify U.S. citizenship. Whether the Commerce Department accepts that arrangement, and how quickly, is what the intelligence community's operational picture now turns on.
Originally reported by nytimes.com
Read the original article →Original headline: NYT: NSA Lost Operational Access to Anthropic Mythos When Export Ban Hit, Creating Direct Intelligence Capability Gap