OpenAI says NATO used up to 300 billion tokens a month of its cyber defense models as allied agencies report bugs found

AI & Compute

NATO Consumes Up to 300 Billion Tokens a Month of OpenAI Cyber Models

NATO has used up to 300 billion tokens a month of OpenAI cyber models via NCIA's Daybreak program, as ENISA and CERT Polska report vulnerabilities found and patched.

By
Grace Kim
Filed
Channel
AI & Compute
Read
2 min read

NATO has consumed up to 300 billion raw input and output tokens a month of OpenAI's cybersecurity models, OpenAI said on Tuesday, Nov. 17. The company did not name the models in use, say when access began or disclose what it costs.

The alliance reaches the models through the NATO Communications and Information Agency (NCIA) via Daybreak, OpenAI's program for vetted defenders. Authorized work under Daybreak includes reviewing code, investigating suspicious activity, validating vulnerabilities and testing fixes. The announcement coincided with the opening of NATO Edge 26, NCIA's industry conference in Izmir, Turkey.

NCIA General Manager Dylan Browne called the technology a turning point. "AI-powered cyber defence models represent a generational leap in cyber security," Browne said. He said NATO will study token consumption, usage patterns and the underlying economics of AI models as use scales. In his words, "this insight will help us optimise costs, evaluate competing solutions and make informed procurement decisions."

What have allied agencies actually found?

OpenAI published concrete results from European agencies holding Daybreak access:

  • ENISA, the EU's cybersecurity agency, used OpenAI models to find four vulnerabilities in software used across EU institutions. Fixes have been published for all four.
  • In a separate review of 27 software projects, ENISA and CERT-EU used several frontier models, including OpenAI's. Maintainers judged 95% of 88 potential issues to be genuine.
  • CERT Polska used OpenAI models to help find six vulnerabilities in MikroTik's RouterOS. Researchers directed the work and checked findings on real systems. MikroTik has patched the flaws, and its fixes stopped attacks CERT Polska had observed.

The CERT Polska result is the sharpest commercial signal in the announcement. A major router vendor shipped patches after AI-assisted discovery, and the fixes blocked attacks already observed in the field.

Who else has access?

National cyber agencies in France, Germany and Poland hold Daybreak access. Ukraine's Ministry of Digital Transformation uses the program to defend civilian infrastructure. OpenAI also supported NATO's Crossed Swords 2026 exercise, run by the Cooperative Cyber Defence Centre of Excellence in Tallinn, where participants built and tested AI agents for cyber operations under human oversight.

The program is expanding on the product side as well. OpenAI added two Daybreak agent products last week, Red Team and Defense Factory.

What comes next?

OpenAI faces at least one declared competitor for government cyber work. Google has said it will give Japan early access to frontier models for cyber defense. Google researchers have also reported attackers using an AI tool that turns security patches into working exploits, a finding that strengthens the case for the defensive programs OpenAI is selling.

Browne's remarks point to a procurement contest ahead. NATO intends to compare token costs across vendors before committing to purchases, and the 300 billion token figure gives rivals a baseline to undercut. How OpenAI prices that volume against Google's government offerings will shape the market for AI-assisted cyber defense across allied governments.

Source: mlq.ai

Share this article:

More from Grace Kim

Grace Kim

Show full bio

Market editor covering people, launches and day-to-day practice in Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.

1,053 articles

Related articles

« Previous article