IC Technology & Surveillance — 2026-07-05

NSA and CISA Testing Anthropic Mythos AI Model for Digital Espionage and Cyber Defense as Five Eyes Warns NATO Allies to Step Up Against AI-Powered Threats

BLUFWashington's toggle on Claude Mythos export access functions as a coercive lever over allied cyber posture heading into the Ankara summit, binding NATO partners to a revocable US corporate dependency.

The NSA and CISA have been testing Anthropic's Claude Mythos model for digital espionage and cyber defense applications, according to Politico, citing officials briefed on the trials 1. Anthropic expanded its Project Glasswing access program in June to roughly 150 organizations across more than 15 countries, including the EU, after initially limiting evaluation access to a handful of countries such as the UK 12. The Trump administration imposed export controls on Anthropic's most cyber-capable models in early June, barring foreign nationals from using them, then lifted the restrictions on June 30 after an 18-day lapse 2. Separately, the White House has restricted OpenAI's newest model to a small group of approved US companies 1. The Five Eyes intelligence alliance issued a rare warning urging NATO allies to swiftly strengthen defenses against AI-powered cyber threats, according to Politico 1, and ChinaTechNews reported that Claude Mythos surfaced vulnerabilities in classified US systems within hours during a government test 2.

Analysis
Washington's on-again access policy for Claude Mythos now functions as leverage over NATO allies heading into the Ankara summit, with NSA and CISA already running the model for offensive vulnerability discovery while the Five Eyes warning signals allied services see themselves as exposed and under-resourced against AI-enabled threats. Expansion to roughly 150 organizations across more than 15 countries narrows but does not close the gap driving European complaints, since the administration has already shown export controls can be reimposed and reversed within weeks. A single Politico account underlies all reporting, with other outlets amplifying rather than corroborating independently. The June 30 reversal may reflect summit-timed diplomatic accommodation rather than a durable policy shift, leaving allied cyber defense dependent on a US corporate access decision Washington can revoke on short notice.

3 sources

AI security questions loom over NATO summit - Politico
US control of frontier AI looms over NATO summit - ChinaTechNews.com
US control of frontier AI looms over NATO summit - The Next Web