OpenAI is extending access to its cybersecurity tool Daybreak to NATO and several European governments. The tool lets cyber defense teams use OpenAI models to examine software, look for vulnerabilities and patch security gaps.
The rollout comes as public agencies report using the underlying models in security work, from reviewing software projects to protecting civilian infrastructure. It also reflects a growing role for AI in cyber operations, with human oversight remaining part of the described approach.
Daybreak reaches NATO and government teams
The NATO Communications and Information Agency (NCIA) says it is already making heavy use of the models behind Daybreak. NCIA chief Dylan Browne described the development as a generational leap in cyber defense.
During the NATO exercise Crossed Swords 2026, participants deployed AI agents for cyber operations under human oversight for the first time. The exercise shows how AI agents can be incorporated into operational settings while people supervise their use.
Ukraine is using Daybreak to protect civilian infrastructure. France, Germany and Poland have also been granted access, widening the group of European governments that can use the tool.
Reported findings in software and routers
The source describes several examples of AI models helping security teams identify flaws. The EU cybersecurity agency ENISA used the models to find four vulnerabilities in software used by EU institutions.
A joint report with CERT-EU examined 27 software projects. Of the 88 reported issues, 95 percent turned out to be real bugs. That result indicates that most of the reported issues in those reviewed projects were genuine defects, according to the source.
Poland’s CERT Polska found six vulnerabilities in MikroTik routers that were being actively exploited. This example links vulnerability discovery to an immediate security concern: the flaws were not merely theoretical, but were already being used.
What the rollout suggests for cyber defense
Software audits can help teams spot security problems and close gaps. By using models to support this work, cyber defense teams can examine code and software projects for potential issues. Daybreak is presented as a way for those teams to use OpenAI’s models for that process.
The examples point to different kinds of work: identifying vulnerabilities in software used by EU institutions, examining projects alongside CERT-EU, finding exploited router flaws, and protecting civilian infrastructure in Ukraine. The shared thread is applying AI models to security tasks carried out by public agencies and defense teams.
Human oversight is explicitly part of the Crossed Swords 2026 account. As governments gain access to Daybreak, the source provides a concrete illustration of AI agents being deployed for cyber operations with people supervising them, while agencies use the models to investigate software risks.