OpenAI ships agents that pen-test your app, then patch what they find
Daybreak Red Team hunts for real vulnerabilities in web apps, Defense Factory turns those findings into draft pull requests for an engineer to review.
OpenAI announced two new security tools, Daybreak Red Team and Daybreak Defense Factory, at its Cyber at the Frontier event on 12 November.
Red Team runs a team of AI agents against a web app, reading the source code and application context to work out whether a reported issue, from a code scanner, a bug bounty programme or a vulnerability disclosure, is an actual risk rather than noise. It runs in an isolated environment with a review layer called Guardian controlling what the agents can touch, and it’s open by early access request only. Defense Factory picks up from there: agents trace a vulnerability through the codebase and prepare a fix, pulling in findings from Red Team itself or from tools like Wiz, Snyk and Endor Labs, plus its own scheduled scans of connected GitHub repositories. Nothing merges on its own. Every patch lands as a draft pull request for an engineer to review. It’s a preview plugin in ChatGPT for workspaces with Daybreak access.
Why it matters: this is the find-it-then-fix-it loop that security teams already build by hand with separate tools, packaged as one product from the model vendor itself. If the triage step genuinely cuts false positives, that’s the slow part of a vulnerability backlog actually getting faster.
The caveat: both tools are gated, early access for Red Team and a workspace preview for Defense Factory, so there’s no independent read yet on how often the triage gets it right or how good the patches are.