Overview
- OpenAI unveiled Daybreak Red Team and Daybreak Defense Factory at its Cyber at the Frontier event on Thursday, introducing early access for Red Team and a Defense Factory preview plugin for ChatGPT workspaces.
- Daybreak Red Team runs coordinated AI agents that draft attack campaigns from app source code and context, execute tests in isolated environments, and validate whether reported issues are exploitable.
- Daybreak Defense Factory traces confirmed flaws to specific files, synthesizes code fixes and tests, and produces merge-ready GitHub pull requests for engineers to review rather than auto-deploy.
- OpenAI presented GPT6 Cyber benchmark results and internal counts—including reproducing 81 previously unknown high-severity Firefox flaws and validating nearly 1,000 Linux local privilege escalations—but those performance claims come from company testing and await independent verification.
- The tools integrate with scanners and platforms such as Wiz, Snyk, Endor Labs and GitHub and include supervisor and Guardian controls to limit testing scope, a design that could speed remediation for developers while raising operational trust and governance questions for security teams.