OpenAI Provides NATO With Cybersecurity Models Through Daybreak

OpenAI announced access to advanced cybersecurity models for NATO’s core institutions through the Daybreak program. CCDCOE also confirmed the tool’s use in training AI agents during the Crossed Swords 2026 exercise under human supervision.

OpenAI Daybreak at NATO is intended for authorized cyber defense work. On November 17, OpenAI said that, together with the NATO Communications and Information Agency (NCIA), it is providing NATO’s core institutions with access to advanced cybersecurity models through the Daybreak program.

Separate confirmation came from the NATO Cooperative Cyber Defence Centre of Excellence (CCDCOE). The center said that the technical environment of the Crossed Swords 2026 exercise used OpenAI Daybreak for the first time. The exercise also included a new training module for AI agents in cyber operations.

OpenAI Daybreak at NATO and the Crossed Swords Exercise

According to CCDCOE, human supervision was maintained while working with AI agents. However, public information does not specify exactly how the models’ outputs were checked or what tasks the AI agents performed during the exercise.

OpenAI claims that NATO used up to 300 billion input and output tokens per month for this work. This figure was published by the company itself; NCIA has not yet independently confirmed on its own public website the extent of use or the conditions for access to the models.

It is also not publicly known which specific OpenAI models are available, which NATO networks or systems the deployment covers, or whether this involves production use beyond the declared access and exercises. The sources do not confirm any ongoing cyberattack or exploitation of a vulnerability.

Cyber Defense as Part of Collective Defense

NATO considers cyber defense part of its collective defense. The announcement therefore represents public confirmation of the use of OpenAI tools in the Alliance’s defensive cyber work and in the CCDCOE training environment.

OpenAI and its partners speak of using generative AI to find and verify vulnerabilities. However, claims that this work has been accelerated or provides a security benefit are not yet supported by independent effectiveness measurements.

What Is Still Missing

When AI is deployed in a sensitive defense environment, the main undisclosed issues include data-handling rules, the boundaries of permitted use, and procedures for human approval of outputs. Although CCDCOE confirmed human supervision during the exercise, it has not publicly provided technical or procedural details.

Other important information will include a separate statement from NCIA or NATO on the scope of the cooperation, details about the protection of sensitive data, and independently verifiable deployment results. These could include, for example, data on vulnerabilities removed or an audit of the effectiveness of the procedures used.

Sources

  • OpenAI – Confirms the announcement of cooperation with NCIA, NATO’s declared access to models through the Daybreak program, and the claim about usage volume.
  • NATO Cooperative Cyber Defence Centre of Excellence – Independently confirms the use of OpenAI Daybreak in the technical environment of the Crossed Swords 2026 exercise, the training of AI agents, and the requirement for human supervision.
  • NATO – Describes the role of cyber defense in the Alliance’s defense and the framework of allied cyber capabilities.

Verified and updated: November 17, 2026 15:24

Sharing