OpenAI said on Tuesday that it is working with the NATO Communications and Information Agency to give NATO's core institutions access to its advanced cybersecurity models, and that NATO has used up to 300 billion raw input and output tokens a month as part of that work.
Access runs through Daybreak, OpenAI's program for vetted defenders doing authorized security work. According to OpenAI, that work covers reviewing software, investigating suspicious activity, validating vulnerabilities and developing and testing fixes.
OpenAI made the partnership public as the NATO Edge 26 conference got under way in Turkey, Quartz reported, citing The Wall Street Journal. The company gave no start date for NATO's use of the models, according to the same report.
What NATO's agency says
Dylan Browne, general manager of the agency known as NCIA, called AI-powered cyber defense "a generational leap in cyber security" in a statement released by OpenAI.
NCIA also plans to study cost as use grows, according to a statement quoted by Help Net Security. "As we scale adoption, we will build a deeper understanding of token consumption, usage patterns and the underlying economics of AI models," the statement said, adding that the insight would help the agency "evaluate competing solutions and make informed procurement decisions."
Sasha Baker of OpenAI said the aim is to help defenders "find and fix vulnerabilities before attackers can exploit them."
Agents in a NATO exercise
OpenAI said it also supported the 2026 Crossed Swords exercise, organized by the NATO Cooperative Cyber Defence Centre of Excellence. The exercise added a training stream in which participants built and evaluated AI agents for cyber operations using Daybreak, with an emphasis on keeping humans in oversight, according to OpenAI.
Findings outside NATO
OpenAI said it has given access to its cyber models to the EU's cyber agency ENISA, France's ANSSI and national agencies in Germany and Poland. Ukraine's Ministry of Digital Transformation uses Daybreak to defend civilian infrastructure, the company said.
ENISA used OpenAI models to find four vulnerabilities in software used across EU institutions, with fixes published for all four, according to OpenAI. Help Net Security reported that ENISA and CERT-EU reviewed 27 software projects using several frontier models, OpenAI's among them, and that maintainers confirmed 95% of 88 potential issues as real.
CERT Polska used OpenAI models to help find six vulnerabilities in MikroTik RouterOS, the operating system on MikroTik's network gear, Help Net Security reported. Its researchers directed the work and checked findings on real systems, and MikroTik released fixes, the outlet said.
Context
The announcement landed on the same day Help Net Security reported that Google had found an AI agent writing and launching exploits with no human in the loop, a sign that attackers are adopting the same kind of tools.
The 300 billion figure is a ceiling, according to OpenAI, which described NATO's usage as "up to" that amount a month rather than an average.
Source: OpenAI, “Strengthening NATO and allied cyber defenses with AI”
Additional sources: Help Net Security, “NATO's AI cyber defense work hits up to 300 billion OpenAI tokens a month”; Quartz, “OpenAI gives NATO access to Daybreak cybersecurity AI models”
