NATO Burns 300 Billion OpenAI Tokens Monthly
OpenAI says NATO's core institutions use up to 300 billion input and output tokens a month through the NATO Communications and Information Agency, reviewing software, investigating suspicious activity and testing fixes for allied networks.
Edited and fact-checked by Stephen Sprague
NATO is tracking its own token use and costs, and the procurement decisions that follow will show what the alliance thinks the models are worth.
NATO's cyber defenders run up to 300 billion OpenAI tokens a month through the alliance's own technology agency. The access runs through the NATO Communications and Information Agency, which delivers technology and cybersecurity services to NATO's core institutions, and the work is defensive and authorised, covering code review, vulnerability checks, incident investigation and fix testing.
"AI-powered cyber defence models represent a generational leap in cyber security," said Dr. Dylan Browne, general manager of the NCIA, and he added that the models accelerate vulnerability discovery and detect emerging threats at machine speed. Browne also said the agency is studying its own token consumption and usage patterns, because a meter running at this scale becomes a procurement question as much as a security one.
OpenAI also supported the 2026 Crossed Swords exercise, run by NATO's Cooperative Cyber Defence Centre of Excellence, where participants built, deployed and evaluated AI agents for cyber operations with human oversight kept in place. Exercises are where the alliance can test what these agents do wrong before a real network depends on them, and OpenAI announced the NATO usage figure on November 17.
Europe's civilian agencies have receipts from the same programme, and the European Union Agency for Cybersecurity used the models to find four vulnerabilities in software used across EU institutions, with fixes published for all four. A joint report by ENISA and CERT-EU found that software maintainers confirmed 95 percent of 88 potential issues the models assessed as genuine, and Poland's CERT Polska used them to find six vulnerabilities in MikroTik RouterOS, which the vendor then fixed.
The headline number needs its limits stated plainly, because 300 billion tokens measures processing volume and says nothing on its own about attacks stopped or networks made safer. OpenAI hasn't published the cost of the arrangement or a count of operational results at NATO, and the figure comes from the company selling the service, so readers should treat it as a usage claim awaiting an audit.
The company recently introduced GPT-6.2 Astra for professional work and GPT-6.2 Sol for everyday tasks, and its head of national security policy, Sasha Baker, said OpenAI intends to deepen its work with NATO and allied governments. For NATO's defenders, the practical question is whether the fixes arrive faster than the flaws, and that answer sits in patch records rather than token totals.
The conversation.
Loading comments…
A little more perspective.
Reading is free. Verified members can comment and reply for $2 a month.