
News
OpenAI says NATO is already running its cyber models
OpenAI says NATO’s tech agency is using its cyber models, at up to 300 billion tokens a month, and names European agencies already on Daybreak.
OpenAI disclosed on November 17, 2026 that NATO is already using its cybersecurity models. In a Global Affairs post, the company said it works with NATO’s Communications and Information Agency (NCIA) to give NATO’s core institutions access to advanced cybersecurity models for protecting Allied networks. It said NATO has used up to 300 billion raw input and output tokens a month in that work. OpenAI calls the use early, and writes it as work already underway. Neither that post nor the same-day Wall Street Journal report describes a contract signed that morning.
The Journal, in a story by Mauro Orru timestamped 6:13 a.m. ET, said OpenAI was making the models available to the alliance so it could protect its networks. Orru described NCIA as the alliance’s technology and cyber hub, with a team of over 3,000 civilian and military experts, and said the agency was using the models to probe suspicious activity, detect vulnerabilities, and deploy fixes. That lead does not include the token figure. The 300 billion number is OpenAI’s claim.
OpenAI routes the access through Daybreak. The company says Daybreak puts advanced AI in the hands of trusted defenders for authorized security work: reviewing software, investigating suspicious activity, validating vulnerabilities, and developing and testing fixes. The Journal’s subhead adds that government agencies and enterprise customers can access advanced models for cybersecurity work under the program.
NCIA General Manager Dr Dylan Browne, quoted by OpenAI, called AI-powered cyber defence models “a generational leap in cyber security.” He said they accelerate vulnerability discovery, detect emerging threats at machine speed, and help defenders stay ahead of adversaries. He also said that as NATO scales adoption it will need a deeper understanding of token consumption, usage patterns, and the economics of the models, so it can optimize costs, evaluate competing solutions, and make procurement decisions. Sasha Baker, OpenAI’s head of national security policy, said Daybreak is meant to help NATO’s cyber defenders “find and fix vulnerabilities before attackers can exploit them.”
Institutional access and the token figure sit apart from a training event OpenAI also described. The company said it supported the 2026 Crossed Swords exercise, organized by the NATO Cooperative Cyber Defence Centre of Excellence. The exercise included a new training stream in which participants built, deployed, and evaluated AI agents for cyber operations, using Daybreak in the technical environment. OpenAI said participants worked alongside the agents as operators, with an emphasis on maintaining human oversight. The post does not publish a procedure for those agents, and it does not say the exercise is how NCIA spends the monthly tokens.
OpenAI said it remains committed to NATO’s principles for responsible AI adoption and to solutions that are safe and secure. It did not say when NCIA access began, or what share of the token volume is live defense rather than testing.
Beyond NATO, OpenAI said it has already given cyber-model access to the EU’s ENISA and to national cyber agencies in France (ANSSI), Germany, and Poland. It did not name the German agency. It then attributed specific defensive results to users of the models:
- ENISA identified four vulnerabilities in software used across EU institutions, with fixes published for CVE-2026-73432, CVE-2026-73431, CVE-2026-73405, and CVE-2026-73374.
- A November joint report from ENISA and CERT-EU, which OpenAI cites, reviewed 27 software projects using frontier models, including OpenAI models accessed through Daybreak and tools including the Codex security plug-in. Across the models tested, software maintainers confirmed 95% of 88 assessed potential issues as genuine.
- CERT Polska used OpenAI models to help identify six vulnerabilities in MikroTik RouterOS and coordinated disclosure with the vendor. Researchers directed the work, tested findings on real systems, and verified impact. OpenAI said the models accelerated analysis. MikroTik released fixes that CERT Polska confirmed prevented the attacks it had observed.
- Ukraine’s Ministry of Digital Transformation is using Daybreak to strengthen cyber defence of civilian infrastructure, identifying software vulnerabilities and developing and testing fixes more quickly amid persistent cyber attacks.
The 95% figure is for the models in that joint review, not an OpenAI-only score. The MikroTik account, as OpenAI tells it, kept researchers in charge of testing and verification.
Read OpenAI’s November 17 post beside the Journal’s same-day report, and keep the token figure and the Crossed Swords training stream in separate columns. The next public number worth watching is the usage accounting Browne said NCIA still has to build.