
NATO is using OpenAI’s cybersecurity models at massive scale to help defend allied networks, marking a real shift from testing to day-to-day operations.
Story Snapshot
- NATO’s tech arm is working with OpenAI to give core institutions access to cybersecurity models.
- OpenAI says NATO’s usage has reached up to 300 billion tokens per month.
- Approved tasks include code review, suspicious activity checks, vulnerability validation, and fix testing.
- NATO cyber training used OpenAI tools to build and test agents with human oversight.
NATO Works With OpenAI Through Its Communications and Information Agency
OpenAI announced that it works with NATO’s Communications and Information Agency to provide NATO’s core institutions with access to advanced cybersecurity models. The company says the aim is to help protect allied networks by giving vetted defenders tools that can speed routine analysis and remediation. The access comes through Daybreak, which OpenAI describes as a trusted program for approved cyber defenders. United Kingdom Defence Journal coverage echoed the arrangement and its stated purpose.
OpenAI stated that NATO’s use has reached up to 300 billion raw input and output tokens each month as part of this work. The figure points to heavy operational activity, not just a trial. Token counts reflect model interactions across many tasks. The announcement did not tie the number to specific outcomes such as incidents blocked or hours saved. Still, the scale suggests NATO analysts are leaning on the system for frequent, high-volume support.
What NATO Is Using The Models To Do
OpenAI listed approved uses that match core defensive needs. Those include reviewing software, investigating suspicious activity, validating found vulnerabilities, and developing and testing fixes in controlled ways. That set aligns with daily chores that often slow defenders. Automating first passes on code review and triage can help teams focus on the highest risks. The description frames the models as assistants under human control, not independent decision makers on live networks.
OpenAI also linked the deployment to training. The company supported the 2026 Crossed Swords exercise with a new track. In that track, participants built, deployed, and evaluated artificial intelligence agents for cyber operations, with an emphasis on human oversight. The exercise setting let defenders try agent workflows while keeping people in charge of actions and approvals. The account did not describe exact permissions or whether agents could change live systems.
Why This Matters For Citizens Who Worry About Digital Risk And Waste
Allied networks face constant probes from criminal groups and hostile states. Faster code review and better triage can cut the time it takes to find and fix weak spots. Many readers, right and left, say government tools feel slow and outdated. This move shows a major alliance trying to modernize its defenses. It aims to use artificial intelligence to reduce noise, surface real threats sooner, and help human teams act before attackers can do damage.
Governments worldwide are shifting from artificial intelligence experiments to real operations in cyber defense. A 2026 survey of national security experts found many expect artificial intelligence to improve tasks like vulnerability discovery and incident response. Yet more than half also believed attackers gain faster in the near term, which keeps pressure on defenders to adapt. That tension explains why institutions are racing to test and adopt credible tools that can raise the baseline.
Known Limits In The Public Record
The available public information centers on OpenAI’s announcement and matching press reports. Those sources confirm access, use cases, and the token scale. They do not share contract terms, model versions, start dates, covered networks, or measured outcomes. Cybersecurity often restricts what can be shared, which can limit outside review. The core, uncontested facts remain that NATO’s Communications and Information Agency is using OpenAI’s cybersecurity models through Daybreak, at reported large scale.
NATO has used up to 300 billion OpenAI tokens a month on cyber defense.
Daybreak gives the alliance's core institutions access to advanced cybersecurity models.https://t.co/FilvIuqCRc— The Circuitry (@thecircuitry_) November 18, 2026
Independent policy voices are urging stronger evaluation of advanced models used in government. Analysts argue that public institutions need third-party checks that match the claims vendors make, so leaders and taxpayers can see value and risk in plain terms. United States policy guidance has also called for audits that test fitness for purpose and controls. Those ideas point to a path where adoption and accountability grow together over time.
What To Watch Next
Watch for NATO or member states to share high-level performance metrics that do not reveal sensitive details. Examples include time-to-triage, time-to-fix, or reduced backlogs. Look for follow-on training that keeps humans in charge while using agents to cut routine work. Also track whether allied cyber teams expand use beyond review and validation to guided remediation, with checks that prevent new bugs. These signals will show if the capability is sticking and delivering real gains.
Sources:
insiderpaper.com, openai.com, defence-industry.eu













