OpenAI launches Daybreak Red Team and Daybreak Defense Factory
Confirmed
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- 3 of 5
- Last verified
- 17 November 2026
Your AI and this story
- GPT-6.2 Astra165 days after its cutoff
- Claude Opus 5.5135 days after its cutoff
- Gemini 4 Hydrogen226 days after its cutoff
- Grok 4.7165 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 135 days before it.
Key facts
- Daybreak Red Team: agents use source code and app context to probe attack paths; also validates findings from scanners, bug bounties and disclosure reports and tests the patches; early access
- Nancy Wang (lead, cyber applied AI): ‘superhuman pen-test at the scale of automated continuous scanning’
- Daybreak Defense Factory: agents on OpenAI’s latest cyber models (currently GPT-6 Cyber) scan codebases and write patches; engineers approve before production; Wang: engineers ‘get a merge-ready pull request with the patch and a test that shows that it works’
- Brockman: ‘you need to really reorient your security practice around what these models can do’
- No pricing was given in The Register’s report; Help Net Security and SecurityWeek covered the launch on Nov 13, and EdTech Innovation Hub reported a Hacktron partnership for testing connected systems
- NATO (Nov 17, OpenAI post and Help Net Security; WSJ headline): NATO’s Communications and Information Agency (NCIA) gives NATO’s core institutions access to OpenAI’s cyber models through Daybreak, using up to 300 billion tokens a month; NCIA general manager Dylan Browne said it will ‘build a deeper understanding of token consumption’. Also listed: ENISA found four vulnerabilities in EU-institution software and, with CERT-EU, reviewed 27 projects where maintainers confirmed 95% of 88 flagged issues; CERT Polska found six MikroTik RouterOS flaws; Ukraine’s digital ministry, France’s ANSSI and German and Polish agencies use Daybreak; NATO’s Crossed Swords 2026 exercise built AI cyber agents in it
What happened
Daybreak started in September as OpenAI’s programme for defenders (subsidised access, the Daybreak Defense Network and tiered cyber models). The Nov 12 products turn it into packaged services: one for offensive testing of a customer’s own applications, one for continuous defensive patching. Both rely on the cyber-tuned models OpenAI restricts to vetted users.
Why it matters
Frontier labs are now selling autonomous offensive testing directly, a capability they also list among their main risks. The same week, a coalition of nonprofits sued five labs to freeze frontier development, citing AI-driven cyber incidents.
Sources
5 sources from 3 sites. Numbers match the chips in the text.
5 sources: 3 primary, 2 press
Primary
- OpenAI: Daybreak Red Teamopenai.com, official
- OpenAI: Daybreak Defense Factoryopenai.com, official
- OpenAI: Strengthening NATO and allied cyber defenses with AI (Nov 17)openai.com, official
Press
Changes
- NATO/NCIA and European agencies on Daybreak (news agent)
- Filed (news agent)