NATO Uses OpenAI Models to Strengthen Cyber Defence
By Suad Seferi ·

Key takeaways
- NATO is using OpenAI cybersecurity models. Through cooperation with NCIA, advanced AI systems support vulnerability analysis, threat investigation and security operations across allied networks.
- Usage has reached up to 300 billion tokens monthly. The figure reported by OpenAI indicates substantial AI processing activity, although it does not measure cybersecurity effectiveness.
- AI is entering European defence and security operations. NATO exercises and projects involving European cybersecurity agencies demonstrate growing practical adoption, with human oversight remaining essential.
NATO is using advanced artificial intelligence models developed by OpenAI to strengthen its cyber defences, investigate security threats and identify vulnerabilities across allied networks. OpenAI confirmed the cooperation with the NATO Communications and Information Agency (NCIA), the organisation responsible for delivering technology and cybersecurity services to NATO's core institutions. According to the company, NATO has used up to 300 billion AI tokens per month as part of its cybersecurity operations, indicating substantial use of AI within the alliance's digital security infrastructure. The announcement provides one of the clearest examples of advanced AI models being incorporated into the cybersecurity operations of a major international defence organisation. How NATO is using artificial intelligence The cooperation operates through OpenAI's Daybreak initiative, which provides advanced AI capabilities to authorised cybersecurity professionals. The technology supports several tasks, including reviewing software for vulnerabilities, investigating suspicious activity, analysing potential threats and developing and testing security fixes. These activities traditionally require significant time and specialised expertise. AI models can assist security teams by examining large amounts of technical information and identifying potential weaknesses for further investigation. However, the technology does not eliminate the need for human expertise. Security professionals remain responsible for validating findings, assessing risks and deciding which corrective actions to take. NCIA General Manager Dr Dylan Browne described AI-powered cyber defence as a major advancement in cybersecurity, arguing that the technology can accelerate vulnerability discovery and help defenders respond to emerging threats. Browne also identified another priority: understanding the financial and operational implications of using advanced AI models at scale. As NATO expands its use of these systems, it intends to assess token consumption, operating costs and the performance of competing solutions to inform future procurement decisions. The reported 300 billion tokens represent AI processing volume, not the number of cyberattacks detected or prevented. OpenAI has not publicly disclosed the financial terms of the cooperation or detailed performance results from NATO's operational deployments. AI agents are already being tested in NATO cyber exercises The cooperation extends beyond routine cybersecurity analysis. OpenAI also participated in Crossed Swords 2026, a cyber operations exercise organised by the NATO Cooperative Cyber Defence Centre of Excellence in Tallinn, Estonia. The exercise introduced a dedicated training component in which participants designed, deployed and evaluated AI agents for cybersecurity operations. These systems were tested alongside human operators, allowing participants to examine how AI could assist with complex security tasks while maintaining human oversight. The exercise brought together more than 250 participants from over 40 NATO and non-NATO countries. Its organisers confirmed that OpenAI Daybreak was used in the technical environment, providing independent confirmation of the technology's involvement in the exercise. The distinction between training and operational deployment remains important. The exercise demonstrates how AI agents are being evaluated in cyber operations, but does not establish that autonomous agents have been given unrestricted control over NATO's live security infrastructure. European cybersecurity agencies are also adopting AI OpenAI's cooperation with NATO forms part of a broader effort to introduce advanced AI models into government cybersecurity operations. The company reports that its models have also been made available to European cybersecurity organisations, including the European Union Agency for Cybersecurity (ENISA) and national agencies in France, Germany and Poland. OpenAI says ENISA used its models to identify four vulnerabilities in software used by European Union institutions, with fixes subsequently published. In Poland, CERT Polska used OpenAI models to assist research that identified six vulnerabilities in MikroTik RouterOS. Ukraine's Ministry of Digital Transformation is also using Daybreak to support cybersecurity work involving civilian infrastructure. These examples suggest that AI-assisted vulnerability research is moving beyond experimental demonstrations into practical security activities. Nevertheless, the reported results concern specific projects and do not establish how effectively the same technology performs across every cybersecurity environment. What this means for the Western Balkans The development is relevant to NATO members in Southeast Europe, including Albania, Montenegro and North Macedonia. As members of the alliance, these countries participate in a collective security framework that increasingly depends on shared digital infrastructure, cybersecurity cooperation and advanced technological capabilities. NATO's adoption of AI-assisted cybersecurity tools could influence future technical standards, training programmes and operational practices across allied institutions. However, OpenAI's announcement does not establish that these models have been deployed within the national cybersecurity systems of individual Balkan countries. For governments in the region, the development raises practical questions about cybersecurity readiness, access to advanced technology and the expertise required to operate AI-assisted security systems responsibly. It also introduces questions about dependence on commercial AI providers, the handling of sensitive information and how security organisations independently validate AI-generated findings. These considerations become particularly important when AI systems are introduced into defence networks and other critical infrastructure. The next challenge is measuring results NATO's use of OpenAI te…
Frequently asked questions
How is NATO using OpenAI technology?
NATO uses advanced cybersecurity models to assist with vulnerability identification, threat investigation, software analysis and the development and testing of security fixes.
What does 300 billion AI tokens per month mean?
Tokens are units of information processed by AI models. The figure represents reported processing volume, including input and output tokens. It does not indicate how many cyberattacks were detected or prevented.
Are AI agents operating autonomously inside NATO?
AI agents have been tested during NATO-related cyber exercises. However, the available announcements do not establish that autonomous AI agents have unrestricted operational control over NATO's live cybersecurity infrastructure.
Does this cooperation affect North Macedonia?
North Macedonia is a NATO member and participates in the alliance's collective security framework. The cooperation is relevant to future cybersecurity capabilities, but no specific deployment within North Macedonia's national systems has been confirmed.