Field report NW-18 / P6 · sheet /deals

M&A & DealsFiled from the territory

OpenAI Gives NATO Its Daybreak Cybersecurity AI Models

NATO is pushing up to 300 billion tokens a month through OpenAI's Daybreak cyber models via NCIA, the company said Tuesday, with access already extended to ENISA and agencies in France, Germany and Poland.

Grid
NW-18 / P6
Surveyed
November 20, 2026
Reading
3 min read
Correspondent
Priya Raman
OpenAI is giving NATO access to its Daybreak cybersecurity AI models
Fig. NW-18 / P6OpenAI is giving NATO access to its Daybreak cybersecurity AI models — Pixelkult / pixabay

Field notes

  1. NATO has used up to 300 billion input and output tokens per month through NCIA, OpenAI said Tuesday.
  2. OpenAI disclosed the partnership as the NATO Edge 26 conference began in Turkey, per The Wall Street Journal.
  3. Daybreak split in August into Daybreak Blue (defense, incident response) and Daybreak Red (exploit validation, security testing).
  4. Access extends beyond NATO to ENISA and national cyber agencies in France, Germany and Poland.
  5. NCIA employs more than 3,000 civilian and military experts, the Journal reported.

NATO is now running up to 300 billion tokens per month through OpenAI's cybersecurity models, the company confirmed on Tuesday, in a partnership that puts commercial AI at the center of allied network defense.

Tokens are the units of data AI models process and generate. Three hundred billion of them a month, in raw input and output combined, signals deployment at industrial scale — not a pilot. The figure covers cyber defense work flowing through NCIA, NATO's Communications and Information Agency, which serves as the alliance's hub for technology and cybersecurity and employs more than 3,000 civilian and military experts, according to The Wall Street Journal.

What is Daybreak, exactly?

Daybreak is OpenAI's program for giving authorized defenders access to its cyber models. The toolset lets them examine code, investigate odd network behavior, confirm weaknesses, and build and trial patches. OpenAI gave no start date for NATO's use of the models, the Journal reported.

The company widened the program in August with two access tiers and a model built specifically for security researchers. The structure now splits into:

  • Daybreak Blue — defensive work such as vulnerability discovery and incident response
  • Daybreak Red — exploit validation and security testing

Why the timing matters

OpenAI made the partnership public as the NATO Edge 26 conference in Turkey got under way, the Journal reported. The announcement landed alongside another data point: OpenAI backed the 2026 Crossed Swords exercise, run by the NATO Cooperative Cyber Defence Centre of Excellence.

That exercise added a new training stream. Participants designed, launched and assessed AI agents for cyber operations built on Daybreak — then worked as operators alongside the agents. OpenAI said human oversight was central to the training. The message is deliberate. The agents do the machine-speed work; people stay in the loop.

What are NATO's leaders saying?

Dylan Browne, general manager of NCIA, framed the arrangement as a learning exercise in AI economics as much as a defense tool. "As we scale adoption, we will build a deeper understanding of token consumption, usage patterns and the underlying economics of AI models," Browne said. "This insight will help us optimise costs, evaluate competing solutions and make informed procurement decisions."

That quote does a lot of work. It tells you NATO hasn't settled its long-term AI vendor strategy yet. Token consumption is a cost line, and Browne is openly saying the alliance intends to use real usage data to benchmark competing solutions before committing to procurement.

Sasha Baker, OpenAI's head of national security policy, kept the framing on defense. Daybreak is meant to help NATO's cyber defenders "find and fix vulnerabilities before attackers can exploit them," Baker said. OpenAI added that it looks forward to deepening the work with NATO and its allies.

How far does the access spread?

NATO is the headline, but not the whole story. OpenAI said it has already extended access to its cyber models to defenders across Europe, including:

  • ENISA, the European Union's cybersecurity agency
  • National cyber agencies in France, Germany and Poland

That footprint matters for strategy. OpenAI is positioning Daybreak as the default AI layer for Western state cyber defense — one alliance, one EU agency, three national agencies, and growing.

The commercial logic underneath

For OpenAI, the deal answers a pressing question: where does sustained, high-volume token demand come from? Government cyber defense offers exactly that. Three hundred billion tokens a month from a single institutional customer is the kind of anchor usage that makes infrastructure investment defensible.

For NATO, the calculus runs the other way. Attackers are automating. Defenders who analyze code and patches by hand are losing that race. Daybreak's pitch — examine code, spot anomalies, confirm weaknesses, trial fixes — maps directly onto the bottleneck tasks in security operations centers.

Neither side disclosed financial terms or contract length. What's on the record is the scale, the two-tier product structure, the human-oversight training model, and Browne's stated intent to treat the data as procurement leverage. That last part may prove the most consequential. If NATO publishes what it learns about token economics, every government shopping for cyber AI gets a benchmark for free.

via qz.com (Source)

Share this article:

More from Priya Raman

Priya Raman

Show full bio

Senior reporter covering technology and product news in corporate strategy and chief-executive leadership at CEO Frontier.

496 articles

Filed under

Adjacent sheets

« Previous articleNext article »

End of sheetNW-18 / P6 · November 20, 2026