NATO uses OpenAI models to protect alliance networks as cyber defenders train with AI agents

By Lukasz Prus (Defence Industry Europe)

NATO |
NATO uses OpenAI models to protect alliance networks as cyber defenders train with AI agents

Image: NATO.

NATO has used up to 300 billion input and output tokens a month in its cybersecurity work with OpenAI, the company said. OpenAI is working with the NATO Communications and Information Agency to give the alliance’s core institutions access to AI models that can help protect their networks.

Dylan Browne, the agency’s general manager, said the models could help NATO find vulnerabilities faster and spot emerging threats. As their use grows, the agency plans to examine how much computing work the models handle and what that costs.

“As we scale adoption, we will build a deeper understanding of token consumption, usage patterns and the underlying economics of AI models,” Browne said. “This insight will help us optimise costs, evaluate competing solutions and make informed procurement decisions.”

OpenAI’s Daybreak gives authorized cyber defenders tools to review software, investigate suspicious activity, verify vulnerabilities and test fixes. “Through Daybreak, we’re putting more capable tools in their hands to help them find and fix vulnerabilities before attackers can exploit them,” said Sasha Baker, OpenAI’s head of national security policy.

At the 2026 Crossed Swords exercise, participants used Daybreak to build, deploy and evaluate AI agents for cyber operations. The exercise, organized by the NATO Cooperative Cyber Defence Centre of Excellence, added a training stream in which participants worked alongside the agents and retained human oversight.

OpenAI said it has also given the EU’s cybersecurity agency, ENISA, and national cyber agencies in France, Germany and Poland access to its models. In Ukraine, the Ministry of Digital Transformation is using Daybreak to help teams find software flaws and develop and test fixes for systems that support essential civilian services.

ENISA used OpenAI’s models to identify four vulnerabilities in software used across EU institutions, and fixes were published for all four, the company said. A separate November report by ENISA and CERT-EU reviewed 27 software projects using several AI models, including OpenAI models accessed through Daybreak, and tools including the Codex security plug-in.

Across the models tested, software maintainers confirmed that 95% of the 88 potential issues they assessed were genuine. In Poland, CERT Polska used OpenAI’s models to help identify six vulnerabilities in MikroTik RouterOS and coordinated disclosure with the vendor.

CERT Polska’s researchers directed the work and tested their findings on real systems, using AI to speed up the analysis. MikroTik issued fixes that CERT Polska confirmed prevented the attacks it had observed.