The TL;DR
US control over the most powerful cyber AI models, led by Anthropic’s Claude Mythos, is over the NATO summit in Ankara on 7-8 July. Washington has stepped in on export controls and expanded joint access through Project Glasswing, frustrating its European allies who want access while building their own defensive AI. By law, the conference will not discuss it.
Donald Trump arrives at next week’s NATO summit in Ankara wielding unusual power, as the US decides which allies get access to the world’s most advanced AI, Politico reports. The consortium meets on 7 and 8 July with AI security questions hovering over the agenda.
A new wave of models from Anthropic and OpenAI can detect and exploit security flaws better than most human experts. Claude Mythos of Anthropic encountered a vulnerability in classified US systems within hours during a government audit.
“AI is fundamentally changing the threat landscape, and NATO needs to adapt accordingly,” Estonia’s cyber ambassador Helen Popp told Politico. All the powers available to the enemies are also available to the allies, counter, if they go first.
💜 for EU tech
The latest talk from the EU tech scene, a story from our genius founder Boris, and some incredible AI art. It’s free, every week, in your inbox. Register now!
US agencies including the NSA and CISA have been investigating Mythos for cyber security and digital espionage. European partners have complained about access, and EU institutions have openly sought it, as only a few countries, including the UK, were initially allowed to carry out the tests.
Anthropic expanded its Project Glasswing program in June to about 150 organizations in more than 15 countries, including the EU. The controversy followed weeks of whiplash from Washington.
In early June, the Trump administration imposed export controls on Anthropic’s internet-intensive models, barring foreigners from using them and enforcing a global shutdown. The controls were removed on June 30 after an 18-day power outage.
The White House also limited the release of the latest model of OpenAI to a small group of authorized US companies, per Politico. The push and pull has frustrated allies, prompted a rare Five Eyes warning about AI cyber threats, and left-wing models moving between governments faster than regulators can keep up.
Quiet corridors, noisy subtitles
The conference agenda includes a track on emerging and disruptive technologies, but an official told Politico that AI and cyber will get a brief mention in the closing remarks. Former NATO cyber policy chief Heli Tiirmaa-Klaar said the allies avoided formally discussing controversial topics, predicting talks on the fringes instead.
The US State Department’s cyber bureau is not sending a representative amid an internal reorganization, Politico reports. Senator Jeanne Shaheen said she would go some distance to assure her allies that the US will not “divide” them with access to AI models.
Trump separately signed NSPM-11, which ordered the US military to quickly deploy AI and protect models from China. Europe is attractive by building its own forces, including the AI defense alliance between Helsing and Mistral.
The war in Ukraine, now in its fourth year, remains concrete, and allies have pledged 1.5% of GDP to protect critical infrastructure. Laura Galante of the Center for European Policy Analysis called Ukraine a blueprint for AI-fueled warfare.
A State Department spokesman said every alliance must use “trusted advanced AI capabilities”. What skills count as trust, and who gives trust, is exactly what Ankara will not really discuss.
