OpenAI says it is working with NATO's Communications and Information Agency to give NATO institutions access to advanced cybersecurity models through Daybreak. According to the company, NATO has used up to 300 billion raw input and output tokens in a month, and the 2026 Crossed Swords exercise included a stream in which participants built and evaluated agents for cyber operations with human oversight.
The company also describes work with Ukraine's Ministry of Digital Transformation, ENISA and national cyber agencies. It says ENISA used frontier models in a review of 27 software projects and that maintainers confirmed 95% of 88 potential issues assessed across the models tested. CERT Polska, meanwhile, is reported to have used OpenAI models in work that identified six RouterOS vulnerabilities subsequently fixed by MikroTik. Those outcomes are attributed to OpenAI and its linked partners; they should not be treated as a general accuracy rate for AI vulnerability research.
The scale of use makes governance more important than the headline capability. Authorised access, scope control, human validation, disclosure coordination and cost measurement all become operational requirements. NATO's own procurement and oversight experience could therefore be as valuable to enterprise defenders as the vulnerability findings themselves.
