NATO’s cyber agency used up to 300B OpenAI tokens a month

NATO’s Communications and Information Agency (NCIA) has used up to 300 billion raw input and output tokens per month of OpenAI’s cybersecurity models to help protect Allied networks, OpenAI said in a November 17 post. The same post lists results outside NATO. CERT Polska used OpenAI models to help find six vulnerabilities in MikroTik RouterOS, and MikroTik has released fixes that CERT Polska confirmed prevented the attacks it had observed.
The access runs through Daybreak, OpenAI’s program that gives trusted defenders advanced AI for authorized security work: reviewing software, investigating suspicious activity, validating vulnerabilities, and developing and testing fixes. Through NCIA, it reaches NATO’s core institutions.
Bugs found, fixes shipped
In the Polish case, researchers directed the work, tested findings on real systems and verified their impact, while the models sped up analysis and the search for potential vulnerabilities. CERT Polska coordinated the disclosure with MikroTik, so operators of affected RouterOS devices now have a concrete way to protect them.
ENISA, the EU’s cyber agency, used OpenAI models to identify four vulnerabilities in software used across EU institutions, and fixes have been published for all four. A November joint report from ENISA and CERT-EU goes wider. It describes a review of 27 software projects using frontier models, including OpenAI models accessed through Daybreak, along with tools such as the Codex security plug-in. Across the models tested, 95% of the 88 potential issues that software maintainers assessed were confirmed as genuine.
Ukraine’s Ministry of Digital Transformation, whose access opened in September for civilian infrastructure, now has a stated job for the tool: helping Ukrainian teams identify software vulnerabilities and develop and test fixes more quickly. OpenAI says it has also given its cyber models to national cyber agencies in France (ANSSI), Germany and Poland.
What NATO is measuring
OpenAI also supported the 2026 Crossed Swords exercise, run by the NATO Cooperative Cyber Defence Centre of Excellence (CCDCOE). A new training stream there had participants build, deploy and evaluate AI agents for cyber operations using Daybreak. They worked alongside the agents as operators, with the emphasis on keeping humans in oversight.
Dr Dylan Browne, NCIA’s general manager, called AI-powered cyber defence models “a generational leap in cyber security” that “accelerate vulnerability discovery” and “detect emerging threats at machine speed.” Sasha Baker, OpenAI’s head of national security policy, said Daybreak puts more capable tools in defenders’ hands “to help them find and fix vulnerabilities before attackers can exploit them.”
OpenAI is not alone in selling this kind of tool. It launched Daybreak earlier in 2026, not long after Anthropic released its cyber-focused model Mythos. And Browne said NCIA will be keeping count as its use grows: “As we scale adoption, we will build a deeper understanding of token consumption, usage patterns and the underlying economics of AI models. This insight will help us optimise costs, evaluate competing solutions and make informed procurement decisions.”
Up next
Claude Haiku 5.5 spotted a data leak GPT-6.1 Sol missed
GPT-6.1 Sol saves $0.77 per task against Claude Haiku 5.5 in the sudo L7 benchmark, yet it called a copied feature ready to deploy though it would expose one client’s data to another. Has a coding agent ever told you code was ready when it wasn’t?
Related stories
- OpenAI’s Daybreak Red and Blue land in Amazon Bedrock
- OpenAI opens Daybreak to Ukraine’s civilian cyber defense
- Agent approvals join ZDR in OpenAI’s enterprise pitch
- Cloudflare checks AI-found code flaws against live traffic
- Sophos lets Daybreak agents close 52% of MDR cases
- Defense Factory sets out OpenAI’s defender playbook
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

