On Nov. 17, Dataconomy reported that OpenAI said it was working with NATO’s Communications and Information Agency (NCIA) to give NATO core institutions access to cybersecurity models through Daybreak. El Periódico separately reported that NATO was already using OpenAI models. The reported work is aimed at authorized cybersecurity tasks, not general-purpose access. (Dataconomy; El Periódico
What Daybreak is intended to do
OpenAI described the work as identifying software weaknesses, validating vulnerabilities, and developing and testing fixes, according to Dataconomy. In practical terms, that takes the work from finding a weakness through checking it and testing a repair.
Dataconomy describes Daybreak’s program rules as limiting testing to systems a user owns, operates, or has explicit permission to assess. The rules also bar sharing access with external users, and national-security or intelligence use requires review and approval under OpenAI policies.
Authorization is the hinge: Daybreak’s described security work is for systems the user owns, operates, or has permission to assess.
The NATO Cooperative Cyber Defence Centre of Excellence said Daybreak was used in the technical environment for Crossed Swords, where participants built and evaluated AI agents for cyber operations. The exercise organizer emphasized human oversight, Dataconomy reported.
California’s separate lawsuit
El Periódico reported that California filed a lawsuit alleging OpenAI prioritized growth and rapid model development over safety and the public interest after a cyber incident involving Hugging Face. The allegations concern that incident and are distinct from the reported NATO access arrangement.