NATO adopts OpenAI's Daybreak: NCIA consumes up to 300 billion tokens per month
On November 17, 2026, NATO and OpenAI jointly announced that the alliance is implementing Daybreak, OpenAI's cybersecurity platform launched in May of the same year.
It is a clear signal that commercial AI models built for cybersecurity are now running at operational scale inside NATO — but the headline number is an upper bound, the costs are undisclosed, and nearly every claim traces back to a single company announcement with a marketing agenda.
What was announced
At the NATO Edge conference in Izmir, Turkey, NATO and OpenAI announced that NATO has implemented Daybreak, the platform OpenAI launched in May 2026. NATO Communications and Information Agency (NCIA) is using, according to the company, up to 300 billion raw input and output tokens per month (Resilience Media).
Sasha Baker, OpenAI's head of national security policy, said according to BigGo Finance that Daybreak puts more capable tools in NATO cyber defenders' hands to help them find and fix vulnerabilities before attackers can exploit them (BigGo Finance). At the conference, she also highlighted a goal of reducing response times to attempted or active breaches from months "to minutes" — Resilience Media notes that this is a promise, or at best an aim, not a documented result.
NCIA general manager Dylan Browne called AI-driven cyber defence models "a generational leap in cybersecurity" (BigGo Finance) — a company claim, not a verified fact. Browne also said NCIA will build a deeper understanding of token consumption, usage patterns and the underlying economics in order to optimize costs and make informed procurement decisions (Defence Industry Europe).
Evidence beyond the press release
The most concrete material is not the token figure, but results from European cyber agencies that have used OpenAI's models:
ENISA. The EU's cybersecurity agency used OpenAI's models to identify four vulnerabilities in software in use across EU institutions, and fixes were published for all four, according to Defence Industry Europe, citing the company. In a separate November report, ENISA and CERT-EU reviewed 27 software projects using multiple AI models, including OpenAI models via Daybreak, and tools such as the Codex security plug-in. Maintainers confirmed 95% of the 88 assessed findings as genuine (Defence Industry Europe).
CERT Polska. The Polish agency used OpenAI's models to identify six vulnerabilities in MikroTik RouterOS with coordinated disclosure to the vendor, and MikroTik's fixes were confirmed effective (Defence Industry Europe). Combining two of them gave full control of the device without authentication if the SSH interface was reachable from the internet, and real attacks exploiting the combination were recorded (ForkLog).
Crossed Swords 2026: AI agents in exercise
Crossed Swords 2026, organized by NATO CCDCOE, took place November 2–6 in Tallinn with more than 250 participants from over 40 countries (ForkLog, BigGo Finance). The exercise, which mixed cyber operations with kinetic military scenarios, added a training stream in which participants built, deployed and evaluated AI agents for cyber operations, with operators working alongside the agents and an emphasis on humans in the loop. Exercise leader Josef Kostelansky described, according to ForkLog, one task as practicing human–AI interaction while maintaining human control over automated systems.
A broader footprint
OpenAI says access has been granted to ENISA and national cyber agencies in France, Germany and Poland. In Ukraine, the Ministry of Digital Transformation is using Daybreak to help teams find software bugs and develop and test fixes for systems supporting essential civilian services (Defence Industry Europe).
The economics — and the gaps
Resilience Media estimates that the token bill could start at around $60,000 per month for input alone, based on an apparent price of $20 per 1 million input tokens for OpenAI's cyber models. Output tokens cost, by the same calculation, $30 to $100 per 1 million, which could push costs considerably higher (Resilience Media). This is Resilience Media's own back-of-envelope calculation — not a disclosed contract value.
According to the analysis firm Konsulteer, OpenAI has not disclosed which specific models are used, the financial value of the arrangement, or how the monthly token consumption splits between input and output (Konsulteer).
Open questions
Nearly all claims originate from a single OpenAI announcement relayed by secondary sources. The 300 billion token figure is an upper bound of raw input and output combined, not average usage. The figure of 220 identified potential vulnerabilities in the ENISA–CERT-EU study is verified only in a single AI-translated source and conflicts with other outlets that mention only 88 assessed findings. And whether NATO actually pays OpenAI — and if so, how much — is undisclosed.
Worth watching going forward: cost transparency, whether competing vendors enter the picture, and whether the ENISA–CERT-EU methodology is published in full — that would provide an independent basis for judging whether AI findings in European cyber defence actually hold up.

