OpenAI announced that the North Atlantic Treaty Organization's technology agency will get access to its Daybreak cybersecurity models, after a first outing in a Tallinn exercise.

OpenAI said on Nov. 17 that the North Atlantic Treaty Organization's Communications and Information Agency will get access to its advanced cybersecurity models through Daybreak, the company's program for approved security work

2

1

. It is a notable expansion of the company's national-security work.

Source: Quartz

Source: Quartz

The arrangement puts OpenAI's models in the hands of the alliance's core institutions for defensive cyber work

2

1

. OpenAI describes the uses as finding software weaknesses, validating vulnerabilities, building and testing fixes, examining code and investigating unusual network behavior

2

1

.

Source: Benzinga

Source: Benzinga

Per the Wall Street Journal, the agency OpenAI is working with is the alliance's hub for technology and cybersecurity, employing more than 3,000 civilian and military experts

1

. The same report puts the announcement alongside the alliance's Edge 26 conference in Turkey

1

.

Scale and the alliance's own framing

The alliance's cyber work already runs on a lot of OpenAI computing power. The agency uses as many as 300 billion tokens a month across input and output, per OpenAI's own figures

2

1

.

OpenAI's new customer, for its part, frames the deal as a way to spend smarter. The agency's general manager, Dylan Browne, said: "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"

1

.

A first outing in an alliance exercise

Daybreak saw its first use in an alliance exercise earlier this month. The Cooperative Cyber Defence Centre of Excellence's Crossed Swords exercise in Tallinn, which ran Nov. 2 to 6, put Daybreak in its technical environment

2

1

. Participants built AI agents for cyber operations there and evaluated them

2

1

. The exercise's organizer emphasized that the agents worked under human oversight

2

1

.

The tiers and the guardrails

OpenAI expanded the program in August into Daybreak Blue, for defensive work such as finding vulnerabilities and handling incidents, plus Daybreak Red, for exploit validation and security testing

2

1

.

The program's access rules require approval and restrict use to systems a customer owns or is authorized to test

2

. OpenAI's policies also prohibit malicious cyber activity and require review for national-security or intelligence use, per the company's own access guide

2

.

A growing government roster

OpenAI has named other Daybreak recipients: the EU's cybersecurity agency, its French counterpart, national cyber agencies in Germany and Poland, and Ukraine's Ministry of Digital Transformation, which uses the program to protect civilian infrastructure

2

1

.

Per a November report by EU cyber agencies, Daybreak users have access to the GPT-5.5 and GPT-5.6 families, Astra, and Codex with its security plug-in

2

. The same report said its reviews examined source code, not running applications, and that the models could misjudge how severe a vulnerability is without seeing it deployed

2

. Poland's computer emergency response team said its researchers drew on OpenAI models to help find six vulnerabilities in MikroTik RouterOS, though each still had to be verified on real systems

2

.

Per Quartz, OpenAI's program launched earlier in 2026 in competition with Anthropic's Project Glasswing, with original partners including Cisco, CrowdStrike, Cloudflare and Palo Alto Networks

1

.

The alliance joins that customer list, but OpenAI has not said which models it uses, when the deal began, what it costs, or which tier it sits on

2

1

.

© 2026 TheOutpost.AI All rights reserved