OpenAI said on November 17, 2026 that NATO's core institutions will get access to its most advanced cybersecurity models through Daybreak, the company's program for vetted cyber defenders. According to Help Net Security, the access runs through the NATO Communications and Information Agency, known as NCIA, and is meant to help protect Allied networks. (Help Net Security (opens in a new tab); The Decoder (opens in a new tab))

Defense One reported that OpenAI executives made the announcement at NATO Edge in Izmir, Turkey, the alliance's largest annual technology business event, which drew more than 3,500 attendees and more than 270 exhibitors. (Defense One (opens in a new tab))

What Daybreak is

Daybreak launched in May 2026. Defense One described it as a program that gives accredited institutions and the U.S. government access to OpenAI's most capable models for finding vulnerabilities in important software. The publication noted that Anthropic runs a similar effort, Project Glasswing, which launched in April. (Defense One (opens in a new tab))

Help Net Security said access is limited to vetted defenders doing authorized work, such as reviewing code, investigating suspicious activity and testing fixes, with people directing the work and checking the results. (Help Net Security (opens in a new tab))

NATO's usage of the models reaches up to 300 billion raw input and output tokens per month, according to Help Net Security. NCIA general manager Dylan Browne said the agency will build a deeper understanding of token consumption as adoption grows. The Decoder reported that Browne described Daybreak as a generational leap in cyber defense. (Help Net Security (opens in a new tab); The Decoder (opens in a new tab))

Results from European agencies

OpenAI also named several European users. The EU cybersecurity agency ENISA used the models to find four vulnerabilities in software used across EU institutions, and fixes are available for all four, according to Help Net Security. (Help Net Security (opens in a new tab); The Decoder (opens in a new tab))

In a joint review, ENISA and CERT-EU examined 27 software projects with several frontier models, including OpenAI's. Maintainers assessed 88 potential issues and confirmed 95 percent of them as real bugs. (Help Net Security (opens in a new tab); The Decoder (opens in a new tab))

Poland's CERT Polska used the models to help find six vulnerabilities in MikroTik RouterOS, the software that runs MikroTik routers. Help Net Security reported that its researchers directed the work and checked the findings on real systems, and that MikroTik released fixes which CERT Polska confirmed stop attacks it had observed. (Help Net Security (opens in a new tab); The Decoder (opens in a new tab))

Ukraine's Ministry of Digital Transformation uses Daybreak to defend civilian infrastructure, and national agencies in France, Germany and Poland have also been given access, including France's ANSSI. (Help Net Security (opens in a new tab); The Decoder (opens in a new tab))

AI agents in a NATO exercise

During NATO's Crossed Swords cyber exercise in Tallinn, Estonia, earlier in November, participants built and deployed AI agents for cyber operations and worked alongside them as operators. Defense One cited NATO's announcement saying the exercise used on-premises AI and, for the first time, OpenAI Daybreak. Help Net Security said the exercise put an emphasis on keeping humans in control. (Help Net Security (opens in a new tab); Defense One (opens in a new tab))

Sasha Baker, who leads national security and policy at OpenAI, said the goal is to shrink the time between the discovery of a vulnerability and its patch from months to minutes, according to Defense One. A NATO official told the conference that NCIA will award more than $3 billion in contracts to industry by the end of 2027, with a focus on cloud computing, space services and AI. (Defense One (opens in a new tab))

The announcement shows how quickly frontier AI labs are becoming suppliers to national security agencies. Both OpenAI and Anthropic now run programs that hand their strongest models to vetted defenders, on the argument that defenders need the same tools attackers will soon have. The European results, with most reported issues confirmed as real and fixes shipped, give that argument concrete numbers that go beyond marketing claims.