News
November 18, 2026 · OpenAI

OpenAI extends its Daybreak security tool to NATO and European governments

OpenAI is opening Daybreak, which lets cyber defenders use its models to audit code and fix vulnerabilities, to NATO and governments including Germany, France and Poland.

OpenAI said on November 17, 2026 that it is working with NATO and allied cyber agencies to give trusted defenders access to Daybreak, its program for using AI models in security work. According to The Decoder's report on the announcement, Germany, France and Poland are among the governments that now have access, alongside NATO itself.

What's new

  • What Daybreak does: The Decoder describes Daybreak as a way for defensive security teams to put OpenAI's models to work checking code, spotting weaknesses and fixing them.
  • NATO use: NATO's communications and information agency (NCIA) says its teams are already heavy users of the models behind Daybreak, The Decoder reports. The alliance's Crossed Swords 2026 exercise marked the first time participants let AI agents carry out cyber operations, with humans supervising.
  • Earlier results: according to the same report, ENISA, the EU's cybersecurity agency, turned up four flaws with the models in software that EU bodies rely on. In a joint exercise with CERT-EU that looked at 27 software projects, 95 percent of the 88 flagged issues were confirmed as genuine bugs.
  • Other users: The Decoder also says Poland's national CERT uncovered six flaws in MikroTik routers that attackers were already exploiting, and that Ukraine relies on Daybreak to defend civilian systems.

Context

In its separate GPT-6.2 launch, OpenAI said its newest models will also be offered through Daybreak, which gives vetted security researchers access with fewer guardrails than the public products, The Decoder notes.

Why it matters

Analysis: AI labs increasingly treat strong security abilities as both a product and a risk. Routing the least restricted access through a vetted program for defenders, while tightening limits for everyone else, is the pattern OpenAI is now applying to government agencies. Security teams elsewhere should expect similar programs, and stricter refusals in general-purpose tools such as ChatGPT.

Source: OpenAI