OpenAI gives NATO institutions access to AI models for finding and fixing cyber flaws
NATO has already used up to 300 billion input and output tokens a month, OpenAI says. Officials want faster repairs—and a clearer picture of the economics.
Loading page…
NATO has already used up to 300 billion input and output tokens a month, OpenAI says. Officials want faster repairs—and a clearer picture of the economics.
Listen to this story
OpenAI says Daybreak usage has reached as much as 300 billion input and output tokens a month as NATO institutions and European public-sector defenders use the program for authorized cyber-defense work. Reported examples include flaws found and patched, but the results are agency-specific and do not establish NATO-wide impact or an OpenAI-only detection rate. NATO’s communications agency says it still needs cost and usage data for procurement decisions; OpenAI’s goal of cutting repair timelines from months to minutes remains an aspiration, not a demonstrated NATO outcome.
Daybreak supports software review, suspicious-activity investigations, vulnerability validation, and development and testing of fixes for trusted defenders.
ENISA says it found four vulnerabilities in software used across EU institutions, and fixes have been published for all four.
In a joint report covering 27 software projects, maintainers confirmed 95% of 88 assessed potential issues as genuine; several models were tested, not OpenAI alone.
NATO’s core institutions have access to OpenAI models for finding software weaknesses and testing repairs, with usage already reaching up to 300 billion raw input and output tokens a month, the company said November 17. OpenAI’s announcement puts a scale on cyber-defense work delivered through Daybreak, its program for trusted defenders conducting authorized security work.
The NATO Communications and Information Agency, or NCIA, provides the access to help protect Allied networks. OpenAI executives discussed the announcement at NATO Edge in Izmir, Turkey, according to Defense One. The arrangement concerns cybersecurity models for defenders, rather than unrestricted access for any purpose.
This is an international extension of an existing effort, not Daybreak’s debut. The program launched in May, giving accredited institutions and the U.S. government access to advanced models for finding software vulnerabilities. The new disclosure describes both NATO’s institutional access and work already underway across European cyber agencies.
Daybreak’s stated scope covers several stages of defensive work: reviewing software, investigating suspicious activity, checking whether suspected weaknesses are real, and developing and testing fixes. That connects discovery with repair, rather than limiting the program to producing possible bug reports. OpenAI describes the access as intended for trusted defenders performing authorized work.
OpenAI also identified government users outside NATO’s core institutions. Their participation spans EU-level software security, national cyber defense and the protection of civilian services:
The clearest examples in OpenAI’s announcement involve specific flaws and published repairs. The company says ENISA used its models to identify four vulnerabilities in software used across EU institutions. Fixes have been published for all four. Those are agency results cited by OpenAI, not a measured improvement across NATO’s networks.
OpenAI also pointed to a November joint report from ENISA and CERT-EU covering 27 software projects. It used several advanced models, including OpenAI models accessed through Daybreak, and tools including the Codex security plug-in. Software maintainers confirmed 95% of the 88 potential issues they assessed as genuine. That figure covers the models tested together; it is not an OpenAI-only success rate.
In Poland, CERT Polska used OpenAI models to help find six vulnerabilities in MikroTik RouterOS. Researchers directed the work, tested findings on real systems and verified their impact, OpenAI said. MikroTik released fixes, and CERT Polska confirmed that they prevented the attacks it had observed. Here, the evidence extended beyond identifying a weakness to checking that a repair worked.
Human oversight also featured in NATO’s 2026 Crossed Swords exercise. OpenAI supported a new training stream in which participants built, deployed and evaluated AI agents for cyber operations, using Daybreak in the technical environment. Participants worked alongside the agents as operators. The exercise offered operational training, while emphasizing that people remained responsible for oversight.
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.
Dr Dylan Browne, NCIA general manager, in OpenAI’s announcement
Browne’s comments put cost control and comparisons with competing tools on the procurement agenda as adoption scales. The monthly token figure describes consumption, while his statement identifies a separate task: understanding usage well enough to make buying decisions. It does not attach a price or a savings figure to that volume.
At NATO Edge, Sasha Baker, OpenAI’s head of national security policy, set out a different measure of success: shortening the interval between discovering a vulnerability and patching it. She described moving from months to weeks, hours and minutes as the outcome the company wants to achieve—not a result it has already demonstrated for NATO.
Baker explicitly invited NATO’s audience to hold OpenAI accountable for that goal. Her proposed test is faster repairs, not simply more model use.
Loading discussion...
Join the conversation
Explain how you would balance stronger defenses against public spending.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.