The NATO Communications and Information Agency (NCIA) has integrated OpenAI’s cybersecurity models into the core infrastructure of the alliance, resulting in a monthly consumption of up to 300 billion raw input and output tokens. This deployment is facilitated through Daybreak, a specialized program designed by OpenAI to provide vetted cybersecurity professionals with access to advanced models for authorized defensive tasks. These activities include the systematic review of source code, the investigation of suspicious network activity, and the development and testing of security patches.
The scope of this initiative extends beyond NATO’s internal operations to include several European institutions and national cyber agencies. The European Union Agency for Cybersecurity (ENISA) and CERT-EU have utilized these models to analyze 27 distinct software projects. In the course of this work, researchers evaluated 88 potential security issues, confirming that 95 percent of them represented genuine vulnerabilities. ENISA has confirmed that it successfully identified and remediated four specific vulnerabilities within software utilized across EU institutions.
National cybersecurity agencies are also leveraging the technology for defensive purposes. France’s ANSSI, along with counterparts in Germany and Poland, have been granted access to the platform. Furthermore, the Ukrainian Ministry of Digital Transformation is currently employing Daybreak to bolster the defense of civilian infrastructure. During the 2026 Crossed Swords exercise, hosted by the NATO Cooperative Cyber Defence Centre of Excellence (CCDCOE), participants utilized the platform to develop and test autonomous agents for cyber operations, with a specific focus on maintaining human oversight throughout the process.
Practical applications of these models have already resulted in tangible security improvements for network hardware. CERT Polska utilized OpenAI’s models to identify six vulnerabilities within MikroTik RouterOS, the operating system powering various network devices. Following the discovery, MikroTik released security updates to address the flaws. CERT Polska has since verified that these patches effectively neutralize the attack vectors they previously observed.
NCIA leadership has indicated that the current focus is on developing a comprehensive understanding of the economics and operational requirements associated with large-scale AI integration. By monitoring token consumption and usage patterns, the agency aims to optimize costs and evaluate the performance of competing AI solutions. This data-driven approach is intended to inform future procurement decisions as the alliance continues to scale its adoption of artificial intelligence for defensive cyber operations.
While the integration of these models has proven effective in identifying and patching vulnerabilities, the reliance on third-party frontier models remains a subject of ongoing strategic evaluation. The agency is prioritizing the optimization of these tools to ensure they remain sustainable and effective for long-term use across Allied networks. The success of these initial deployments, particularly in the context of critical infrastructure protection and software auditing, suggests that AI-assisted defense will remain a central component of NATO’s cybersecurity posture moving forward.





