What changes with the opening

OpenAI has granted NATO and several European governments access to its Daybreak cybersecurity tool, according to The Decoder. Through Daybreak, the source explains, cyberdefense teams can use OpenAI's models to verify software, identify vulnerabilities and close security gaps. Access was granted, in addition to NATO, to Germany, France and Poland; Ukraine is using it to protect civilian infrastructure.

The novelty does not lie in the tool's existence, but in extending its use to institutional defense structures, with direct involvement of the Atlantic alliance. The NATO Communications and Information Agency (Ncia) says it already makes extensive use of the models underlying Daybreak. The agency's head, Dylan Browne, described this development as a generational leap in cyberdefense: this is an attributed statement, not a fact that can be independently verified on the basis of the available material.

The results already reported

According to the article, the European cybersecurity agency Enisa used the models to identify four vulnerabilities in software used by EU institutions. A joint report with CERT-EU examined 27 software projects: 95 percent of the 88 reported problems turned out to be a real defect. This is the most concrete data point for assessing the tool's effectiveness, because it concerns the confirmation rate of the identified problems and not just the number of reports.

In Poland, CERT Polska found six vulnerabilities in MikroTik routers that, according to the source, were already being actively exploited. The information is relevant because it indicates use on systems where a gap is not theoretical but exposed to exploitation. During the NATO Crossed Swords 2026 exercise, moreover, participants used artificial intelligence agents for cyber operations under human supervision for the first time: this too is a statement reported by the source, which does not specify technical methods or evaluation criteria.

Why it matters and what remains to be clarified

The opening to governments and military structures signals a shift of artificial intelligence tools toward operational defense tasks, with a potential impact on public software auditing processes. The comparison between the cited cases shows a common element: the tool is presented as support for human analysis, not as a substitute, both in the NATO exercise and in the work of Enisa and CERT-EU.

However, several open questions remain that the material does not clarify. No financial terms, duration or conditions of the access granted to governments are indicated, nor the type of oversight exercised by the beneficiary entities. It is not known whether the identified vulnerabilities were already known, and no details are provided on the software involved. The 95 percent figure on confirmed problems refers to 88 reports across 27 projects and cannot be extended beyond that sample. The positive assessments from Ncia come from the entity itself and should be compared with independent verifications, which at present do not appear in the available material.

Source and attribution: this original summary was prepared by AICronaca based on the report by The Decoder ↗. The statements are attributed to the source; any informational limitations: The news is based on a single article from The Decoder. Original press releases from OpenAI, NATO, Enisa or CERT-EU are not available, nor are details on contractual terms, duration of access and methodology of the verifications. The numerical data cited refer to specific samples and cannot be generalized. The statements by Dylan Browne and Ncia are attributed and not independently verified..