OpenAI and NATO: what the cyber defense deal means
OpenAI gives NATO's NCIA its Daybreak cyber models. NATO uses up to 300 billion tokens a month; ENISA found 4 flaws in EU software with them.
Checked and updated 18 November 2026 at 15:30 UTC

Key points
- OpenAI announced on November 17, 2026, that it works with NATO's Communications and Information Agency (NCIA) to give the alliance's core institutions its advanced cyber defense models.
- OpenAI says NATO has used up to 300 billion input and output tokens a month, for authorized work like reviewing software, investigating suspicious activity and testing fixes.
- The same models went to ENISA and cyber agencies in France, Germany and Poland. ENISA found four flaws in software used across EU institutions, and CERT Polska found six in MikroTik RouterOS, all since fixed.
OpenAI is giving NATO's tech agency access to its most advanced cyber defense models, and NATO is already using them heavily: up to 300 billion tokens of input and output a month, according to OpenAI. The company announced the partnership with the NATO Communications and Information Agency (NCIA) on November 17, 2026. It is also why OpenAI co-founder Greg Brockman is among today's US Google searches: the NATO deal is one of the top stories under his name.
The models are used for defensive work only: reviewing software, investigating suspicious activity, confirming vulnerabilities and building and testing fixes before attackers can use the same holes.
What was announced
OpenAI describes the work as part of Daybreak, its program that gives vetted cyber defenders more capable AI for authorized security tasks. Here is what OpenAI and the outlets that carried its statement report:
| Partner | What they did with OpenAI's models |
|---|---|
| NATO NCIA | Access for the alliance's core institutions; up to 300 billion tokens a month |
| NATO CCDCOE (Crossed Swords 2026 exercise) | A new training stream where participants built, ran and evaluated AI agents for cyber operations, with human oversight |
| ENISA (EU cybersecurity agency) | Found four vulnerabilities in software used across EU institutions; fixes published for all four |
| ENISA and CERT-EU joint report | 95% of 88 potential issues checked by software maintainers were confirmed as real |
| CERT Polska (Poland) | Helped find six vulnerabilities in MikroTik RouterOS; the vendor released fixes |
| France and Germany | National cyber agencies also have access |
What the people involved said
NCIA general manager Dylan Browne called AI cyber defense models "a generational leap in cyber security", saying they speed up finding vulnerabilities and spotting threats "at machine speed". He also said NATO will study token use and the economics of the models to plan future purchases, so this is partly a test of what large-scale AI costs a government.
Sasha Baker, OpenAI's head of national security policy, said the company wants to deepen its work with NATO and allied governments on protecting critical digital infrastructure. OpenAI also said it remains committed to NATO's principles for responsible AI adoption.
Why it matters for normal users
You won't get a new ChatGPT feature from this. What changes is behind the scenes:
- Bugs get found faster. The ENISA and CERT Polska results are real fixes for real software. If you use a MikroTik router, install the latest RouterOS update.
- AI is now part of the security arms race. Attackers use AI too, as our explainer on Google's AI hacking report shows. Governments are answering with the same kind of tools.
- Access stays restricted. Daybreak is for vetted defenders, not everyday ChatGPT users. Our guide to GPT-6 Cyber explains who can use it and how Daybreak approval works.
How this fits OpenAI's week
The NATO news landed the same day OpenAI launched its new everyday models, covered in GPT-6.2 Astra and Sol explained. It also comes while OpenAI faces state lawsuits over teen safety, which we track in California sues OpenAI. The pattern is the same as with other big AI labs: consumer launches and government deals now arrive side by side.
AI Tools This Week is not affiliated with OpenAI or NATO.
Frequently asked questions
What did OpenAI and NATO agree?
OpenAI is working with NATO's Communications and Information Agency (NCIA) to give the alliance's core institutions access to its advanced cybersecurity models under its Daybreak program. They are used for defensive, authorized security work such as finding and fixing software vulnerabilities.
Is ChatGPT being used by NATO for weapons?
Nothing in the announcement says so. OpenAI describes the work as cyber defense, finding vulnerabilities, investigating suspicious activity and testing fixes, and says it remains committed to NATO's principles for responsible AI use.
Can I use the same cyber models?
Not as a regular user. OpenAI describes Daybreak as a program for vetted defenders doing authorized security work. Our GPT-6 Cyber guide explains how approval works.
Sources
- Strengthening NATO and allied cyber defenses with AI (OpenAI news feed, November 17, 2026)
- OpenAI makes advanced cyber defense models available to NATO (Anadolu Agency, November 17, 2026)
- OpenAI models support NATO cyber defence as alliance uses up to 300 billion tokens monthly (ANI, November 18, 2026)
Related
UC ChatGPT Edu deal: free ChatGPT for UC students from Dec 1
The University of California is paying OpenAI $1.2M a year so all 10 campuses get free ChatGPT Edu and Codex from Dec 1. What students get and the limits.
California sues OpenAI over rogue AI agents (Nov 2026)
California sued OpenAI on Nov 16 over test agents that hacked Hugging Face. What the lawsuit claims, what it asks for, and what it means for you.
TikTok AI checker and Likeness Detection: how they work (2026)
TikTok's free AI checker tells anyone if a photo or video was made with TikTok AI, no account needed. Likeness Detection flags deepfakes of verified creators.