OpenAI launches Daybreak Red Team and Defense Factory for AI vulnerability testing and patching

OpenAI announced Daybreak Red Team and Daybreak Defense Factory at its Cyber at the Frontier event on November 12. Red Team uses multiple AI agents to investigate potential weaknesses in web applications, examining code scanners, bug bounty submissions, and vulnerability disclosure reports. Defense Factory traces findings through code, develops patches, and creates draft GitHub pull requests after engineers review and test proposed changes. OpenAI said patches are not deployed automatically.

The product documentation supplied by OpenAI gives the operational boundary: agents use source code and application context to confirm vulnerabilities, while Guardian review enforces testing boundaries and controls accessible applications and endpoints. Defense Factory supports imports from Wiz, Snyk, and Endor Labs, scans connected GitHub repositories, monitors new commits, and runs scheduled assessments. The evidence trail is a vendor description, not independent testing, and early access is limited to requested users.

This places advanced AI inside the security workflow: discovery, validation, patch drafting, and repository monitoring. The design keeps humans at the merge point, but the agents can move from scanner noise to code-level attack paths. That is an operational shift from advisory AI to tools that shape remediation. The same capability is dual-use: faster validation helps defenders, while the same attack-path logic is valuable to anyone probing software weaknesses.