AI · Nov 9, 2026
OpenAI’s Control Room puts a fail-closed kill switch inside the inference engineOpenAI's cyber models are now part of NATO's defense stack
· Updated · Signal Desk newsroom
Analysis of reporting by OpenAI. How we work

NATO's communications agency has used up to 300 billion tokens a month of OpenAI cyber tooling, and European agencies report confirmed bug finds. It is a real defensive win, and a deepening dependency on one vendor.
Key takeaways
- OpenAI works with NATO's Communications and Information Agency to give NATO's core institutions access to advanced cybersecurity models; NATO has used up to 300 billion tokens a month.
- The 2026 Crossed Swords exercise added a training stream where participants built and evaluated AI agents for cyber operations using Daybreak, with an emphasis on human oversight.
- ENISA used OpenAI models to find four vulnerabilities in software used across EU institutions, with fixes published.
- An ENISA and CERT-EU review of 27 projects found 95% of 88 potential issues assessed by maintainers were confirmed as genuine, across the models tested.
- CERT Polska found six MikroTik RouterOS vulnerabilities with AI help; NCIA says it will weigh costs and competing solutions as use grows.
From pilot to infrastructure
OpenAI's latest Daybreak update reads like a progress report from inside allied cyber defense. The company says it works with NATO's Communications and Information Agency, known as NCIA, to give NATO's core institutions access to advanced cybersecurity models, and that NATO has used up to 300 billion raw input and output tokens per month. That is not a trial. It is a sizable operational workload.
It builds on the program we covered at launch in Daybreak for frontline defenders and its extension to Ukraine's civilian defense. Beyond NATO, OpenAI lists access for the EU's cyber agency ENISA and national agencies in France, Germany and Poland, building on its partnership with Poland's NASK.
The evidence that it works
The most persuasive material comes from the agencies' own outputs. ENISA used OpenAI models to identify four vulnerabilities in software used across EU institutions, and fixes have been published for all four. A November joint report from ENISA and CERT-EU reviewed 27 software projects using frontier models, including OpenAI models via Daybreak and the Codex security plug-in; across the models tested, 95% of 88 potential issues assessed by maintainers were confirmed as genuine. Note that this figure spans several models, not OpenAI's alone.
CERT Polska used OpenAI models to help find six vulnerabilities in MikroTik RouterOS, coordinated disclosure, and confirmed that MikroTik's fixes blocked the attacks it had observed. Researchers directed the work and verified findings on real systems. That human-led pattern, AI accelerating analysis while experts decide, is the right model, and the NATO Crossed Swords exercise reinforced it by training participants to build and evaluate cyber agents with an emphasis on human oversight.
The dependency question
The quote from NCIA's general manager is telling. Alongside praise for machine-speed defense, he says that as adoption scales NATO will study token consumption and the economics of the models to optimize costs, evaluate competing solutions and make procurement decisions. That is a healthy signal that the alliance does not intend to be captured by a single supplier, and it should not be. Military and government cyber defense running on one commercial model provider concentrates risk in pricing, availability and policy decisions made in San Francisco.
There is also the dual-use shadow. The same capability that finds bugs for defenders finds them for attackers, which is why OpenAI gates it through trusted access. WIRED's hosts this week tied a cluster of outages to the strain of rapid AI-driven patching, as we covered in the outage wave. Faster discovery is unquestionably good for defenders. It also means patching discipline and safe rollout have to keep pace.
Who feels it
- Allied cyber agencies
- They gain machine-speed vulnerability discovery but must manage cost and vendor concentration.
- Software vendors
- More AI-found bugs mean more disclosures and more pressure to ship fixes quickly.
- Rival AI labs
- NATO's explicit plan to evaluate competing solutions opens room to compete for defense work.
- Network operators
- Fixes like MikroTik's arrive faster when agencies use AI for discovery.
What to watch
- NATO procurement decisions comparing OpenAI with other model providers.
- Further ENISA and CERT-EU reports on AI-assisted vulnerability discovery.
- Whether GPT-6.2 models are added to Daybreak for allied defenders.
- Disclosure volumes and patch timelines for software used across EU institutions.
FAQ
How is NATO using OpenAI?
OpenAI works with NATO's Communications and Information Agency to give NATO's core institutions access to advanced cybersecurity models, with usage of up to 300 billion tokens a month.
What did ENISA find with AI?
ENISA used OpenAI models to identify four vulnerabilities in software used across EU institutions, all now fixed, and a joint review with CERT-EU found 95% of 88 assessed issues were genuine across the models tested.
What is OpenAI Daybreak?
OpenAI's cybersecurity program that provides vetted defenders with advanced AI capabilities, security tools and support.
Companies: OpenAI
More on OpenAI
AI · Nov 11, 2026
OpenAI’s $2 million for children’s AI safety research matters for one promise: publishing bad newsAI · Nov 17, 2026
OpenAI says its models now shave area off production chips, including AMD’s next flagshipsAI · Nov 3, 2026
OpenAI’s GPT-5.6 Cyber reaches AWS GovCloud through Daybreak Red, with two approval gatesAI · Oct 29, 2026
OpenAI’s new Model Spec swaps rules for five values, trading auditability for judgment