On November 17, 2026, OpenAI announced on its website that it is opening its Daybreak cybersecurity model to NATO and allied cyber defense agencies: NATO's core institutions can now use frontier models directly for authorized security work — reviewing code, investigating suspicious activity, validating vulnerabilities, and developing and testing fixes. NATO already consumes up to 300 billion input and output tokens per month for this work.
How NATO uses it: 300 billion tokens a month, agents trained in exercises
OpenAI works with NATO's Communications and Information Agency (NCIA) to give NATO's core institutions access to advanced cybersecurity models. NATO's early use covers day-to-day protection of Allied networks and helps multinational teams build cyber expertise. The 2026 Crossed Swords exercise added a new training stream in which participants built, deployed, and evaluated AI agents for cyber operations in a technical environment powered by Daybreak, with human oversight throughout. NCIA General Manager Dylan Browne called AI-powered cyber defense models "a generational leap" that can find vulnerabilities and spot emerging threats at machine speed.
Beyond NATO: Ukraine, EU institutions, and Poland are using it too
Beyond NATO, Daybreak has reached several European defense bodies. The EU's cybersecurity agency ENISA used the models to identify four vulnerabilities in software used across EU institutions, with fixes published for all four; a November 6 joint report from ENISA and CERT-EU also describes reviewing 27 software projects with frontier models, where 95% of 88 potential issues were confirmed as genuine by maintainers. Ukraine's Ministry of Digital Transformation uses it to harden the cyber defense of civilian infrastructure, and Poland's CERT Polska used it to find six vulnerabilities in MikroTik RouterOS and coordinated fixes with the vendor.
For defenders, Daybreak matters because it compresses the find-and-fix cycle from weeks to hours — but it goes only to trusted defense institutions, not to ordinary users or companies. At the other end of the same security line, the United States is tightening chip export controls — America closes the AI chip "cloud loophole": renting compute now needs a license too. One side hands allies a stronger shield, the other blocks adversaries from getting compute: the AI security race is shifting from comparing model capability to comparing defense systems.