Security

NATO Uses OpenAI Models to Find Vulnerabilities Across Allied Networks

European defense and cybersecurity agencies are consuming up to 300 billion tokens monthly through OpenAI's Daybreak program to identify and patch security flaws.

Omega Editorial · November 17, 2026 · 3 min read

Key takeaways

  • NATO's cyber agency processes up to 300 billion AI tokens monthly through OpenAI's Daybreak program for vulnerability detection across Allied networks.
  • ENISA and CERT-EU reviewed 27 software projects with AI models and confirmed 95% of 88 flagged security issues as legitimate vulnerabilities.
  • CERT Polska used OpenAI models to discover six vulnerabilities in MikroTik RouterOS, all of which have been patched by the vendor.
  • Ukraine's Ministry of Digital Transformation and national cyber agencies in France, Germany, and Poland now have access to the Daybreak program.
  • NATO's 2026 Crossed Swords exercise tested AI agents for cyber operations while emphasizing the need for human oversight of AI-driven decisions.

NATO's Communications and Information Agency has deployed OpenAI's cybersecurity models across Allied networks, processing up to 300 billion raw input and output tokens per month as defense organizations test AI-assisted vulnerability detection at scale.

Access to the models comes through Daybreak, OpenAI's program designed for authorized security work including code review, threat hunting, and security testing. The program now supports national cyber agencies in France, Germany, Poland, and Ukraine, alongside EU-level institutions.

Confirmed vulnerabilities in production systems

ENISA, the European Union Agency for Cybersecurity, used OpenAI models to identify four vulnerabilities in software deployed across EU institutions, according to details first reported by Help Net Security. All four flaws have been patched.

In a broader assessment, ENISA and CERT-EU reviewed 27 software projects using multiple frontier AI models, including OpenAI's offerings and tools like the Codex security plug-in. Software maintainers evaluated 88 potential security issues flagged by the models and confirmed 95% as legitimate vulnerabilities.

CERT Polska applied the same approach to MikroTik RouterOS, the operating system running on MikroTik network equipment. Researchers directed the AI models and validated findings against live systems, ultimately discovering six vulnerabilities. MikroTik released patches, which CERT Polska verified as effective against observed attack methods.

Operational deployment and cost considerations

Ukraine's Ministry of Digital Transformation is using Daybreak to protect civilian infrastructure. France's ANSSI, the national cybersecurity agency, has also gained access to the program.

At NATO's 2026 Crossed Swords exercise, conducted by the Cooperative Cyber Defence Centre of Excellence, participants built and tested AI agents for cyber operations within Daybreak. The exercise emphasized maintaining human oversight of AI-driven security decisions.

NCIA general manager Dr. Dylan Browne noted that scaling AI adoption will require deeper analysis of token consumption patterns and the underlying economics of AI models. "This insight will help us optimise costs, evaluate competing solutions and make informed procurement decisions," Browne said.

Why it matters

The 300 billion token-per-month figure provides a concrete benchmark for AI consumption in government cybersecurity operations, offering procurement teams and security leaders a reference point for planning large-scale deployments. The 95% confirmation rate for AI-flagged vulnerabilities suggests these models can reduce false positives enough to be practical in production environments, though human validation remains essential. As nation-state actors increasingly target critical infrastructure, the operational deployment by Ukraine's digital ministry demonstrates AI-assisted defense moving from pilot programs to active protection of civilian systems under real threat conditions.

These details were first reported by Help Net Security.

  • #nato
  • #openai
  • #vulnerability detection
  • #cybersecurity
  • #daybreak
  • #enisa

This is an original analysis by the Omega editorial team. Source reporting: AI Watch.

Want systems like this working for your business?

Book a Call

More in Security