OpenAI has detailed a deepening cybersecurity relationship with NATO, revealing that the alliance's institutions are consuming as many as 300 billion raw input and output tokens per month through the company's advanced cybersecurity models. The disclosure, published on OpenAI's blog and picked up by The Wall Street Journal this week, offers one of the clearest pictures yet of how frontier AI is being woven into the cyber defense apparatus of a major intergovernmental military alliance.

How OpenAI and NATO Work Together

The partnership runs through NATO's Communications and Information Agency (NCIA), the technical arm that procures and operates the alliance's networks and command systems. According to OpenAI, the NCIA provides NATO's core institutions with access to advanced cybersecurity models to help protect allied networks.

The company frames the work as a force multiplier for the multinational teams that defend allied networks around the clock. In OpenAI's telling, the tools help strengthen cyber expertise across those teams and generate lessons that can benefit defenders across Europe and the wider alliance — a notable claim given persistent concerns that AI could equally supercharge offensive hacking. The arrangement also lands amid a turbulent stretch for the industry, with breaking AI news dominated by both AI-driven security breakthroughs and market anxiety over frontier-lab finances.

OpenAI says NATO's early adoption shows how capable cyber defenders can put frontier models to work in support of the alliance's core mission of safeguarding the freedom and security of its member states. The company also states that it remains committed to NATO's principles for responsible AI adoption.

What Daybreak Does for Defenders

Central to the story is Daybreak, OpenAI's cybersecurity tool for authorized security work. OpenAI describes defenders using it to review software, investigate suspicious activity, validate vulnerabilities, and develop and test fixes. In other words: the unglamorous, labor-intensive core of defensive security, which AI can compress from days to hours.

Daybreak is not exclusive to NATO. OpenAI says it has already provided access to its cyber models to defenders across Europe, including the EU's cybersecurity agency ENISA and national cyber agencies in France, Germany and Poland. France's ANSSI — one of Europe's most respected national CERTs — is explicitly named, alongside agencies in Germany and Poland.

The Results So Far: Real Vulnerabilities, Real Fixes

The most concrete evidence in OpenAI's disclosure comes from named deployments:

  • ENISA, the EU's cyber agency, used OpenAI's models to identify four vulnerabilities in software used across EU institutions, with fixes published for all four.
  • A November joint report from ENISA and CERT-EU describes a review of 27 software projects using frontier models, including OpenAI models accessed through Daybreak and the Codex security plug-in. Across the models tested, 95 percent of the 88 potential issues assessed by software maintainers were confirmed as genuine — a striking precision rate for AI-flagged security findings.
  • CERT Polska used OpenAI models to help identify six vulnerabilities in MikroTik RouterOS, the operating system running on millions of routers worldwide, and coordinated their disclosure with the vendor. Researchers directed the work, tested findings on real systems, and verified impact. MikroTik released fixes that CERT Polska confirmed prevented the attacks it had observed.
  • Ukraine's Ministry of Digital Transformation is using Daybreak to strengthen the cyber defense of civilian infrastructure, helping teams identify vulnerabilities and develop and test fixes more quickly amid persistent Russian cyberattacks.

AI Agents Enter the Exercise Range

The partnership has also moved into training. OpenAI supported the 2026 Crossed Swords exercise, organized by the NATO Cooperative Cyber Defence Centre of Excellence (CCDCOE) in Tallinn. This year's exercise added a new training stream in which participants built, deployed, and evaluated AI agents for cyber operations using OpenAI Daybreak in the technical environment.

The design is telling: participants worked alongside AI agents as operators, with an explicit emphasis on maintaining human oversight. That mirrors the debate playing out across the security industry about how much authority autonomous agents should have in high-stakes environments, and NATO appears to be deliberately probing the question in a controlled setting rather than leaving it to battlefield improvisation.

Why It Matters

Three currents converge in this story. First, the scale: 300 billion tokens a month is not a pilot program — it reflects routine, heavy use of AI across allied institutions. Second, the direction: the tangible outputs disclosed so far are defensive — identified vulnerabilities and published fixes — in contrast to fears that frontier AI primarily lowers the cost of offense. Third, the precedent: a military alliance institutionalizing frontier AI tooling through a single US vendor raises questions about dependency and sovereignty that European policymakers are already debating in adjacent contexts.

OpenAI says it will continue working with governments, national cyber agencies and international partners to turn frontier AI capability into stronger, more resilient systems. For defenders across the alliance, the message is that AI-assisted defense has moved from experimentation to operational reality — and for the company's critics, the disclosure is a reminder that the same models are now embedded in military-adjacent infrastructure at alliance scale.

Whether NATO's embrace of commercial AI models becomes a template for other alliances, or a cautionary tale about vendor concentration, will depend on results that neither OpenAI nor the alliance has fully quantified yet. What is no longer in question is that the work is underway — and growing fast.

Stay Ahead of AI

For continuous coverage of how governments and alliances are adopting frontier AI, follow AI Buzz Wire for breaking AI news and analysis.

Read more AI news →