OpenAI said on Tuesday that NATO’s Communications and Information Agency gives the alliance’s core institutions access to its cybersecurity models. NATO has used up to 300 billion raw input and output tokens a month in the work, the company said.
What NATO Is Running
Through Daybreak, OpenAI’s program for vetted defenders, the models review software, investigate suspicious activity, validate vulnerabilities and develop and test fixes. The work is authorized security work on Allied networks, and it runs around the clock for NATO’s core institutions.
OpenAI also supported NATO’s 2026 Crossed Swords exercise, run by the Cooperative Cyber Defence Centre of Excellence. A new training stream had participants build, deploy and evaluate AI agents for cyber operations inside Daybreak, with human operators kept in charge.
The Allied Rollout
The EU’s cyber agency, ENISA, used the models to find four vulnerabilities in software used across EU institutions, and fixes are published for all four. A November joint report from ENISA and CERT-EU reviewed 27 software projects using frontier models, including OpenAI’s.
Maintainers confirmed 95 percent of the 88 potential issues assessed as genuine, across the models tested. CERT Polska used the models to find six vulnerabilities in MikroTik RouterOS, and the vendor’s fixes were confirmed to stop the attacks researchers had observed.
Beyond NATO, OpenAI has given access to France’s ANSSI and national cyber agencies in Germany and Poland. Ukraine’s Ministry of Digital Transformation uses Daybreak to defend civilian infrastructure, identifying vulnerabilities and testing fixes amid persistent cyberattacks on essential services that the country faces.
Our Read
Five days after launching Daybreak’s products, OpenAI is pointing to governments as proof of use. The NATO figure is the first hard evidence that a frontier lab’s cyber models have moved from keynote slides into production-scale defense work for a named customer.
It is also the defensive mirror of the attack the industry is already seeing. Google’s threat researchers recently documented AI agents turning software patches into working exploits on their own, and the same capability, in authorized hands, is now defending NATO networks.
The numbers still come from OpenAI alone. No contract values were disclosed, NATO has issued no statement of its own beyond the NCIA quote, and the 95 percent confirmation rate covers several frontier models, not OpenAI’s alone. Treat the token figure as the vendor’s claim, not an audit.
NCIA chief Dylan Browne said the agency will study token use and costs to evaluate competing solutions and make procurement decisions. That matters: NATO is testing the tools, not locked in, and rivals will now pitch the alliance on price, performance and sovereignty.
Why It Matters
Why it matters: government cyber work is becoming a real customer segment for frontier labs, not a hypothetical one. NATO’s adoption gives Daybreak credibility no marketing claim could, and it sets the template for other alliances and agencies to follow.
What to watch is whether NATO turns testing into procurement, and whether the token volume grows beyond 300 billion a month. Also watch the DeepSeek comparison: a free model nearly matched OpenAI’s locked cyber ace in independent testing, so the field is not settled.
Background
Background: OpenAI launched Daybreak Red Team and Defense Factory on November 12, and earlier said it would train on security researchers’ Daybreak sessions. The NATO announcement is the first named government deployment of its cyber models since that launch, five days later.
Source: OpenAI’s NATO announcement and Help Net Security’s report.


0 Comments