NATO uses OpenAI models to protect alliance networks as cyber defenders train with AI agents

Defence Industry Europe

Source published: Site published:

Original article · فارسی

By Lukasz Prus (Defence Industry Europe)

NATO uses OpenAI models to protect alliance networks as cyber defenders train with AI agents

Image: NATO.

NATO has used up to 300 billion input and output tokens each month in its cybersecurity work with OpenAI, the company said. OpenAI is working with the NATO Communications and Information Agency to give the alliance’s main institutions access to AI models that can help keep their computer networks safe.

Dylan Browne, the agency’s general manager, said the models could help NATO find security weaknesses faster and detect new threats. As the agency uses the models more widely, it plans to study how much computing power they use and how much this costs.

“As we use these models more widely, we will learn more about how many tokens they use, how people use them and the costs of running AI models,” Browne said. “This knowledge will help us control costs, compare different solutions and make better decisions when buying technology.”

OpenAI’s Daybreak gives approved cybersecurity teams tools to check software, investigate unusual activity, confirm security weaknesses and test repairs. “Through Daybreak, we’re giving them more powerful tools to help them find and fix security weaknesses before attackers can take advantage of them,” said Sasha Baker, OpenAI’s head of national security policy.

During the 2026 Crossed Swords exercise, participants used Daybreak to create, put into use and test AI agents for cyber operations. The exercise was organized by the NATO Cooperative Cyber Defence Centre of Excellence. It included a new training program in which people worked alongside AI agents while keeping human control over the work.

OpenAI said it has also given its AI models to the EU cybersecurity agency, ENISA, and national cybersecurity agencies in France, Germany and Poland. In Ukraine, the Ministry of Digital Transformation is using Daybreak to help teams find software weaknesses and create and test repairs for systems that provide important services to civilians.

ENISA used OpenAI’s models to find four security weaknesses in software used by EU institutions. The company said that repairs were published for all four. A separate report published in November by ENISA and CERT-EU examined 27 software projects using several AI models. These included OpenAI models accessed through Daybreak, as well as tools such as the Codex security plug-in.

Across all the models tested, software teams confirmed that 95% of the 88 possible problems they reviewed were real security issues. In Poland, CERT Polska used OpenAI’s models to help find six security weaknesses in MikroTik RouterOS. It then worked with the software maker to report the problems.

Researchers at CERT Polska led the work and tested their findings on real computer systems. They used AI to complete the analysis faster. MikroTik released repairs, and CERT Polska confirmed that these fixes stopped the attacks it had observed.