OpenAI Daybreak Powers NATO Cyber Defence

Retro poster illustration of a shield over a map of Europe with a robot, with the words 300 BILLION TOKENS.

On 17 November 2026, OpenAI announced that NATO is using its most advanced cybersecurity AI models at a scale that is hard to picture: up to 300 billion tokens a month. Made at NATO Edge in Izmir, it puts a military alliance at the centre of the AI security story. But the number needs unpacking: 300 billion tokens is a measure of activity, not a count of attacks stopped.

What Daybreak is

Daybreak is OpenAI's programme for putting frontier AI models in trusted defenders' hands. Launched in May 2026, it gives accredited institutions and governments access to OpenAI's most advanced models for authorised security work: reviewing code, investigating suspicious activity, validating vulnerabilities, and testing fixes. The models combine with agent features from Codex and links to corporate security software.

Access is not automatic: OpenAI reviews each request, and approved users may only work on systems they own, operate or are authorised to test. That matters, because the same models that find vulnerabilities could also exploit them — the programme gives defenders the best tools first, under rules keeping them pointed at systems they may touch.

What NATO gets

OpenAI says it works with NATO's Communications and Information Agency (NCIA) to give NATO's core institutions access to its cybersecurity models through Daybreak. The alliance has used up to 300 billion raw input and output tokens a month. NCIA general manager Dylan Browne said the agency will study token use and costs to evaluate competing solutions: NATO is testing OpenAI's tools against rivals before committing.

The announcement does not name the models NATO uses, say when the arrangement began, or disclose financial terms — or whether NATO pays for the access. And there is no NATO or NCIA statement beyond the quote in OpenAI's post, so the figures and framing are OpenAI's: one side's account.

Who else has access

NATO is not the only customer: OpenAI says ENISA, France's ANSSI, and national cyber agencies in Germany and Poland have access, and Ukraine's Ministry of Digital Transformation uses Daybreak to protect civilian infrastructure.

OpenAI also cites results: ENISA found four vulnerabilities in EU-institution software, and all four were fixed. CERT Polska found six in MikroTik RouterOS, and those were fixed too. Those are concrete outcomes — real flaws, found with AI help and closed — and the strongest part of the announcement: a fixed vulnerability is a fact, while a token count is a statistic.

The Crossed Swords exercise

The other half is training. OpenAI supported NATO's 2026 Crossed Swords exercise in Tallinn, 2–6 November, where over 250 participants from 40-plus NATO and non-NATO countries built, deployed and evaluated AI agents for cyber operations. Daybreak was in the technical environment for the first time, alongside locally hosted AI, with participants as operators under human oversight. Sasha Baker, OpenAI's head of national security policy, said Daybreak aims to give trusted defenders tools to identify and fix software vulnerabilities.

What 300 billion tokens does — and does not — prove

A token is a chunk of text, roughly three quarters of a word. Three hundred billion tokens a month is a very large amount of work — code reviews, log analysis, vulnerability checks — at a scale no human team could match. It shows serious adoption by an organisation that does not adopt tools lightly.

What it does not show is results: no count of attacks stopped, vulnerabilities fixed, or incidents prevented. It is not a published audit of month-by-month activity, nor a contract value. The number tells you NATO uses the tools heavily; it does not tell you whether NATO's networks are safer — and OpenAI's announcement does not provide enough public information to calculate that.

Wider context: Anthropic runs a similar programme, Project Glasswing, launched in April 2026. The race to supply defenders with frontier models is now a competition, not a monopoly, and NATO's plan to evaluate competing solutions shows the alliance is shopping, not married to one vendor.

Key takeaways

  • OpenAI announced on 17 November 2026 that NATO's Communications and Information Agency gives NATO's core institutions access to Daybreak cybersecurity models.
  • NATO has used up to 300 billion raw input and output tokens a month — activity, not attacks stopped.
  • Daybreak launched in May 2026 and gives trusted defenders frontier models for authorised security work, with approval-based access.
  • ENISA found four vulnerabilities in EU-institution software and CERT Polska found six in MikroTik RouterOS; all were fixed.
  • Ukraine's Ministry of Digital Transformation uses Daybreak for civilian infrastructure; ANSSI and agencies in Germany and Poland have access too.
  • OpenAI supported the Crossed Swords 2026 exercise in Tallinn, 2–6 November, where 250-plus participants from 40-plus countries tested AI agents.
  • No contract values, model names or start date were disclosed, and there is no NATO statement beyond the quote in OpenAI's post.

Conclusion

Daybreak's NATO deal is a milestone for AI in defence: a military alliance running frontier models at industrial scale to find and fix vulnerabilities, with real fixed flaws to show for it. But the headline number is a measure of use, not of safety. The test that matters — whether these tools make NATO's networks measurably harder to attack — has not been published.

FAQ

What is OpenAI Daybreak?

Daybreak is OpenAI's programme, launched in May 2026, giving trusted defenders access to its most advanced AI models for authorised cybersecurity work — reviewing code, investigating suspicious activity, validating vulnerabilities and testing fixes. Access is approval-based, and users may only work on systems they own, operate or are authorised to test.

How much is NATO using Daybreak?

OpenAI says NATO has used up to 300 billion raw input and output tokens a month through its Communications and Information Agency. It measures activity, not attacks stopped or vulnerabilities fixed.

What vulnerabilities has Daybreak helped find?

OpenAI says ENISA found four vulnerabilities in EU-institution software and CERT Polska found six in MikroTik RouterOS, and all were fixed. The announcement does not name the models used or give contract values.

Sources

Researched with AI assistance and reviewed for accuracy.

Follow on WhatsApp

Follow Daily Tech & AI on WhatsApp for one tech and AI update every day: Daily Tech & AI WhatsApp channel.

 

Comments