OpenAI opens Daybreak cyber tool to NATO and European governments
OpenAI has widened access to Daybreak, its cybersecurity tool, for NATO and several European governments. Early users report confirmed software flaws, including routers already under attack.

Key takeaways
- OpenAI has opened Daybreak to NATO and to several European governments, including Germany, France and Poland.
- ENISA used the models to find four vulnerabilities in software used by EU institutions, in a review of 27 projects.
- CERT Polska found six flaws in MikroTik routers that were already being exploited, according to The Decoder.
- NATO's NCIA says it already uses the underlying models heavily, and AI agents ran cyber operations under human oversight at Crossed Swords 2026.
In this article
What Daybreak does for cyber defence teams
Daybreak lets cyber defence teams point OpenAI's models at software to audit it, search for weaknesses and close the gaps they find. The purpose is defensive: catching flaws before an attacker can make use of them. In practice, a model reads through code and configuration at volume and flags the places that look wrong, and people then check those leads.
The Decoder reports that access has been extended to NATO and to several European governments. Germany, France and Poland are named among the countries granted use of the tool, and Ukraine is already applying it to the protection of civilian infrastructure.
Vulnerabilities found in EU software and MikroTik routers

Two European bodies have already put the models to work. ENISA, the EU's cybersecurity agency, used them to find four vulnerabilities in software that EU institutions rely on. A joint report from ENISA and CERT-EU, the EU's computer emergency response team, covered 27 software projects. Of the 88 issues reported, 95 percent were confirmed as genuine bugs, according to The Decoder.
Poland's national response team, CERT Polska, reported six vulnerabilities in MikroTik routers that were already being exploited. Routers sit at the edge of a network, so a flaw in that equipment is valuable to anyone looking for a way in. The figures suggest the flagged issues are checked by people before they are treated as confirmed.
NATO's cyber agency and the Crossed Swords exercise
The NATO Communications and Information Agency (NCIA) says it is already making heavy use of the underlying models. Its chief, Dylan Browne, described the technology as a generational leap for cyber defence, The Decoder reports.
The alliance has also been putting AI into its training. At Crossed Swords 2026, a NATO cyber exercise, teams ran AI agents through cyber operations with humans supervising, the first time the alliance has done so. That detail matters because it shows where the models sit in practice: inside a chain of command, with people holding responsibility for what the agents do.
Ukraine's use of Daybreak for civilian infrastructure points to the other side of the same work. The same models that help a defence team plan an operation can also help keep everyday services running when they come under pressure.
Who has access and what remains unclear
The practical effect is that a small group of governments now has a quicker way to check its own software. Vulnerability hunting has usually depended on human researchers and scanning tools, and the reported hit rate at ENISA suggests the models can add to that work.
The Decoder's account does not give the terms of access, the cost, or a schedule for other countries. It also does not say whether the ENISA and CERT Polska findings were checked by outside researchers. Agencies that publish their own advisories, and OpenAI's documentation for Daybreak, remain the places to look for technical specifics and for guidance on patching.
Sources
Spotted a mistake? See our corrections policy. Read how News by OnDevice works in our editorial policy.



