OpenAI says NATO has used its cybersecurity models for up to 300 billion tokens a month
OpenAI is offering the models in its Daybreak service to NATO and European cyber authorities. The usage figure shows scale but not results. In OpenAI’s examples, people verified the findings, among them CERT Polska researchers and software maintainers.
OpenAI said on 17 November that it is offering its cybersecurity models to key NATO bodies through the NATO Communications and Information Agency (NCIA). Through the Daybreak service, organisations use the models for authorised security work such as reviewing software and investigating suspicious activity. They also use them to verify vulnerabilities and to develop and test fixes. The Decoder’s report is based on OpenAI’s announcement, so the information comes from the company itself.
OpenAI said NATO has used the models for up to 300 billion tokens a month. A token is a piece of text that a model reads or produces. The figure covers both inputs and outputs. It shows how much the models have been used, not what that use has achieved.
NCIA chief Dylan Browne calls AI a generational leap for cyber defence. Browne said that as use grows, the agency plans to look more closely at token consumption and model costs. It intends to use that information to compare competing solutions and make procurement decisions.
For results, OpenAI points to other organisations. ENISA, the EU’s cybersecurity agency, used the company’s models to find four vulnerabilities in software used by EU institutions, and fixes have been published for all four. OpenAI also cites a joint ENISA and CERT-EU report from November that used several frontier models to review 27 software projects. Software maintainers assessed 88 potential issues and confirmed 95 per cent of them as genuine. That share covers all the models tested, not only OpenAI’s. The Decoder reports the figure without this qualification.
Poland’s CERT Polska used OpenAI’s models to find six vulnerabilities in RouterOS, the software that runs MikroTik routers, OpenAI said. The researchers directed the work, tested the findings on real systems and verified their impact. AI sped up the analysis. MikroTik released fixes, and CERT Polska confirmed that they prevent the attacks it had identified.
At Crossed Swords 2026, an exercise run by NATO’s Cooperative Cyber Defence Centre of Excellence (CCDCOE), a new training module had participants use Daybreak to build and evaluate AI agents for cyber operations. The focus was on human oversight, OpenAI said.
Besides NATO, ENISA, France’s ANSSI and the national cyber authorities of Germany and Poland have been given access. Ukraine’s Ministry of Digital Transformation uses Daybreak to protect civilian infrastructure. The service is intended for trusted defenders. The announcement gives neither its price nor how to apply.
Read more: Models