The OpenAI NATO Daybreak partnership puts OpenAI’s advanced cybersecurity models into the hands of NATO’s core institutions, the company announced on Tuesday, November 17. Through the NATO Communications and Information Agency (NCIA), alliance defenders now use the models for authorized security work, and NATO has already burned through up to 300 billion raw input and output tokens in a single month, according to Help Net Security’s report on OpenAI’s announcement.
The work is defensive, and it is vetted. Daybreak is OpenAI’s program for trusted government and national cyber agency defenders, and the models are used to review software, investigate suspicious activity, validate vulnerabilities, and build and test fixes on allied networks. It is the first time OpenAI has put a number on how heavily a military alliance is using its models: 300 billion tokens a month is production-scale usage, not a pilot.
“Through Daybreak, we’re putting more capable tools in [NATO defenders’] hands to help them find and fix vulnerabilities before attackers can exploit them,” said Sasha Baker, OpenAI’s head of national security policy, who spoke about the work at the NATO Edge conference in Izmir, Turkey. Baker described a goal of cutting the time to respond to an attempted or active breach from months to minutes.
What the Models Have Already Found
The announcement came with concrete results, not just a partnership logo. ENISA, the EU’s cybersecurity agency, used OpenAI models to find four vulnerabilities in software used across EU institutions, and fixes are out for all four. In a separate review, ENISA and CERT-EU examined 27 software projects using several frontier models, OpenAI’s among them, alongside tools such as the Codex security plug-in. Maintainers assessed 88 potential issues and confirmed 95 percent of them as real.

CERT Polska, Poland’s national cyber response team, used OpenAI models to help find six vulnerabilities in MikroTik RouterOS, the operating system on MikroTik’s network gear. Its researchers directed the work and checked the findings on real systems. MikroTik released fixes, and CERT Polska confirmed they stop the attacks it had seen.
Access extends beyond NATO itself. Ukraine’s Ministry of Digital Transformation uses Daybreak to defend civilian infrastructure, and France’s ANSSI and national agencies in Germany and Poland also have access. OpenAI did not disclose what any of these organizations pay for the access.
Training With AI Agents at Crossed Swords
OpenAI also supported NATO’s 2026 Crossed Swords exercise, run by the NATO Cooperative Cyber Defence Centre of Excellence in Estonia. In a new training track, participants built, deployed and evaluated AI agents for cyber operations inside the Daybreak environment, with human operators kept in charge throughout. More than 250 participants from over 40 countries took part in the exercise, according to reports of the event.
“AI-powered cyber defence models represent a generational leap in cyber security,” said Dr Dylan Browne, the NCIA’s general manager. Browne added that as adoption scales, the alliance will build a deeper understanding of token consumption, usage patterns and the underlying economics of AI models, insight he said will help NATO optimize costs, evaluate competing solutions and make informed procurement decisions.

Why It Matters
The partnership marks a shift in who buys frontier AI for security work. Daybreak began as a program for vetted defenders, and OpenAI expanded it last week at its November 12 “Cyber at the Frontier” keynote with Daybreak Red Team and Defense Factory, which GPTOnline covered on November 13. NATO is now its most prominent customer, and the token volume shows the alliance is treating the models as working infrastructure rather than an experiment.
The timing is also hard to ignore. The announcement comes one day after California attorney general Rob Bonta sued OpenAI over autonomous cyberattacks, including the July breach of Hugging Face, arguing the company failed to prevent its models from being used in attacks. OpenAI now finds itself on both sides of the same debate in the same week: accused in court of letting AI agents run loose, and trusted by NATO to help its defenders find flaws before attackers do.
For the alliance, the bet is that the same capabilities that make AI models dangerous in the wrong hands make them valuable in the right ones. The early results, four flaws fixed at EU institutions and six at a major network vendor, suggest the defenders are getting real use out of the tools. The question now is whether the procurement math Browne flagged, token consumption, usage patterns and cost, lets NATO scale that use across the alliance without the bill becoming its own vulnerability.
Sources
- Help Net Security: NATO’s AI cyber defense work hits up to 300 billion OpenAI tokens a month
- Techmeme: OpenAI partners with NATO’s NCIA (Wall Street Journal)
- Defence Industry Europe: NATO uses OpenAI models to protect alliance networks
- Blockchain.News: OpenAI Collaborates with NATO to Enhance Cybersecurity with AI
- GPTOnline: OpenAI Launches Daybreak Red Team and Defense Factory
- GPTOnline: California Sues OpenAI Over Hugging Face Cyberattack

