OpenAI’s allegedly lacklustre approach to security and safeguarding across has just landed the AI giant a major lawsuit in California, but that’s not stopping government entities from working with the company on cyber defence. Today NATO and OpenAI jointly announced that NATO has been implementing Daybreak, OpenAI’s cybersecurity platform that it launched in May 2026. As part of that, NATO’s Communications and Information Agency (NCIA) is using up to 300 billion raw input and output tokens per month.
“The people defending our democracies should have the most advanced technology on their side,” said Sasha Baker, head of national security policy at OpenAI. Speaking at the NATO Edge conference today in Izmir, Turkey, she highlighted a promise (or at least, goal) of going from months down “to minutes” to respond to attempted or active breaches.
How much might this be costing NATO? NATO and OpenAI didn’t disclose details on stage, but at a basic level, OpenAI appears to charge $20 per 1 million tokes for its cyber models, which would work out to a monthly bill of up to $60,000 for tokens alone.
The NATO work has involved testing out models to see how they might be applied across different use cases, as well as training, Baker said.
Alongside this, OpenAI has also disclosed that it is working with NATO on other activities. For example, it supported the NATO 2026 Crossed Swords cyber exercise earlier in November 2026. This included 250 participants from 40 NATO and non-NATO member states working on training in cyber operations and using AI in their “kinetic” work.
OpenAI may be deepening its engagement with NATO, but that work is coming at the same time as the frontier AI company is facing very public scrutiny in its home market over security breaches.
A lawsuit filed in California’s Supreme Court alleges that the company developed and tested advanced artificial intelligence models without any safeguards in place, endangering the public in the process. The suit cites the Hugging Face attack earlier in 2026, where OpenAI’s AI agents, initially trained by OpenAI to operate without human control, broke into Hugging Face’s systems. The suit describes this as one of “at least 80 cyber incidents” where OpenAI knowingly disregarded the risks involved.
“Today my office is suing OpenAI for its development of dangerous AI models without safeguards in place to protect the public,” California’s attorney general Rob Bonta said in a statement. “While narratives about the Hugging Face attack have evoked science fiction fantasies of rogue AI systems run amok, let me be clear: the Hugging Face attack was the predictable result of an internal culture at OpenAI that prioritizes growth and rapid model development over safety and accountability… OpenAI’s actions escalated real-word cybersecurity risks, threatening the safety, health, and well-being of society.” (Read more about AI agents and the implications in national security here.)
Any mention of risks and breaches around OpenAI’s cyber activity was noticeably absent from a panel at NATO Edge on innovation, highlighting the complicated relationship government has with AI right now. Effectively, as Jean Charles Ellermann-Kingombe, the alliance’s assistant secretary general for cyber and digital transformation, noted during the panel, it can pose both the solution as well as the problem. He said this in reference to adversaries using AI, too, but OpenAI’s lawsuit in California highlights that it’s not just adversaries that should invite caution.
For NATO at least for now, it’s choosing to focus on the capabilities. “AI-powered cyber defence models represent a generational leap in cyber security,” noted Dylan Browne, the NCIA’s general manager, in a statement. “They accelerate vulnerability discovery, detect emerging threats at machine speed and enable defenders to stay ahead of adversaries.”
He also sounded a note of wait-and-see, however, when considering how it will engage longer term. “As we scale adoption, we will build a deeper understanding of token consumption, usage patterns and the underlying economics of AI models. This insight will help us optimise costs, evaluate competing solutions and make informed procurement decisions.”
In addition to working with NATO, OpenAI has been ramping up cybersecurity work with individual countries in Europe.
Its cyber models are being used at the EU security agency ENISA, which has reviewed some 27 software projects using Daybreak and other frontier models. France’s (ANSSI), Germany and Poland — which used it to identify six vulnerabilities in MikroTik RouterOS — are also working with Daybreak. Ukraine is implementing Daybreak to identify software vulnerabilities and develop and test fixes more quickly, the company said. supporting defenders who protect essential services amid persistent cyber attacks.












