OpenAI is expanding the use of its advanced cybersecurity models across NATO and European government agencies, highlighting how artificial intelligence is becoming part of software vulnerability discovery, threat investigation and defensive security operations.
The November 17 announcement confirmed that NATO’s Communications and Information Agency (NCIA) provides the alliance’s core institutions with access to AI cybersecurity capabilities through OpenAI’s Daybreak programe.
According to the company, NATO has used up to 300 billion raw input and output tokens per month in the initiative.
The development illustrates growing institutional demand for AI-assisted cybersecurity, but also raises questions about operating costs, accuracy and maintaining human oversight.
NATO Integrates AI Into Cyber Defence
Daybreak gives authorized cybersecurity professionals access to advanced AI models for reviewing software, investigating suspicious activity, identifying vulnerabilities and developing potential fixes.
The NATO partnership also extends to training activities.
During the 2026 Crossed Swords exercise organized by the NATO Cooperative Cyber Defence Centre of Excellence, participants built, deployed and evaluated AI agents for cyber operations.
The exercise emphasized maintaining human supervision while using AI to support technical investigations.
NCIA General Manager Dr Dylan Browne said the technology could accelerate vulnerability discovery and improve the alliance’s ability to protect critical digital infrastructure.
However, Browne also identified the need to understand token consumption, operating costs and the economics of using competing AI systems.
That consideration becomes increasingly important when automated cybersecurity workflows generate substantial computing demand.
European Agencies Report Confirmed Security Findings
The expansion follows AI-assisted vulnerability investigations involving European cybersecurity organizations.
According to OpenAI’s announcement, the European Union Agency for Cybersecurity (ENISA) identified four vulnerabilities in software used across EU institutions, with fixes published for all four.
A separate ENISA and CERT-EU joint assessment examined 27 software projects using frontier AI models and security tools, including OpenAI technologies.
The investigation identified 220 potential issues across the codebases.
Software maintainers had reviewed 88 findings when the report was published, confirming approximately 95% of those assessed as genuine.
The percentage does not represent an overall vulnerability-detection accuracy rate because many findings remained unassessed, and the study does not establish how many vulnerabilities the systems missed.
The findings nevertheless demonstrate how AI-assisted code analysis can support security specialists examining complex software.
AI Models Help Identify Router Vulnerabilities
Poland’s national computer emergency response team, CERT Polska, also used OpenAI models to help identify six vulnerabilities in MikroTik RouterOS.
According to the November 17 disclosure, security researchers directed the investigation, verified the findings on real systems and coordinated remediation with MikroTik.
The vendor subsequently released fixes that CERT Polska confirmed prevented the attacks observed during testing.
Other European organizations receiving access to OpenAI cybersecurity capabilities include France’s ANSSI and government agencies in Germany.
Ukraine’s Ministry of Digital Transformation is also using Daybreak to support the protection of civilian infrastructure.
Cybersecurity Automation Faces a Practical Test
For software engineers and security teams, the developments highlight an opportunity to reduce the time required to investigate vulnerabilities and validate fixes.
However, AI-generated findings still require technical assessment, prioritization and confirmation that recommended patches work correctly.
Automating vulnerability discovery without effective validation could increase review workloads or create misleading security assessments.
The reported investigations demonstrate promising applications, but they do not independently establish how much time or money organizations save through AI adoption.
For developers building cybersecurity products, the larger opportunity lies in combining automated analysis with reliable evidence, human judgment and integration into existing security workflows.
As AI becomes part of critical infrastructure protection, its commercial and operational value will depend on whether it helps defenders identify and resolve genuine security weaknesses more effectively—not simply generate more alerts.