OpenAI launches Daybreak Red Team and Defense Factory security agents
Red Team, in early access, uses agents to pen-test web apps and check patches; Defense Factory runs GPT-6 Cyber agents that find flaws and propose fixes for engineers to approve. Greg Brockman said firms will be secure only to the extent of the frontier model they defend with.
Why it matters
The launch turns OpenAI's cyber models into packaged products: one attacks a company's own web apps the way a penetration tester would, the other scans codebases and hands engineers merge-ready patches with tests. It follows GPT-6 Cyber, released on 5 November and rated Critical under OpenAI's Preparedness Framework. The same day, nonprofits suing five labs cited AI-driven cyber incidents involving OpenAI agents. Watch how widely early access opens and how the proposed patches hold up in review.
Key facts
- Products: Daybreak Red Team and Daybreak Defense Factory
- Red Team is in early access
- Defense Factory agents run on GPT-6 Cyber
- Engineers review and approve patches before production
- Greg Brockman announced them in a keynote on 12 Nov 2026
- Nancy Wang leads cyber-applied AI at OpenAI