THE ESSENTIALS IN 30 SECONDS
- Daybreak Red Team sends coordinated agents against web applications to check whether reported or suspected flaws can actually be exploited, and retests after fixes; it is in early access.
- Daybreak Defense Factory takes findings from Red Team or existing scanners, traces them through the code and proposes patches that developers review, refine with Codex and approve.
- OpenAI has not announced pricing; access to Daybreak requires vetting, and Defense Factory goes through OpenAI's sales team.
OpenAI unveiled two new products for its Daybreak cybersecurity programme at a keynote on 12 November: Daybreak Red Team, which attacks an organisation's web applications to prove which flaws are exploitable, and Daybreak Defense Factory, which turns findings into proposed patches. Both rely on AI agents that read the customer's source code.
Red Team: proving exploitability
According to ITdaily and DigitalToday, Red Team deploys several coordinated agents that combine source code with application context to trace attack paths and confirm whether a vulnerability can really be used. Beyond its own discoveries, it re-examines alerts from code scanners, bug bounty submissions and disclosure reports, so security teams can sort real problems from noise. After a fix ships, it checks whether the exploit still works and can repeat that whenever the application changes. Customers decide which applications and endpoints the agents may touch, and tests run in isolated environments.
Nancy Wang, who leads cyber-applied AI at OpenAI, pitched it as a "superhuman pen-test" running at the scale of continuous scanning, AI Weekly reports. The product is in early access, with organisations invited to register.
Defense Factory: patches under human control
Defense Factory connects to an organisation's repositories and imports findings from Red Team or from existing tools; DigitalToday names Snyk, Endor Labs and Wiz. Agents investigate each issue, follow it through the code and draft a patch. Developers keep the final say: they can inspect the evidence, use OpenAI's Codex coding agent to add context or refine the fix, and open a draft pull request on GitHub. A change is deployed only after a person approves it. The system can scan each new commit or run on a schedule. AI Weekly, citing The Register, says the agents run on GPT-6 Cyber.
The name comes from an approach OpenAI first described internally: an OpenAI page on the "Defense Factory" recounts a security sprint involving more than 250 people in which agents closed 53 urgent or high-priority issues on the first day. Those figures concern OpenAI's own systems, not customer results.
Availability and context
Daybreak access requires an application and vetting by OpenAI. Defense Factory is sold through OpenAI's sales team, and no prices have been announced. At the keynote, president Greg Brockman argued that companies will only be as secure as the frontier model they defend with, a claim that also serves OpenAI's commercial interest.
The launch follows OpenAI's limited release of GPT-6 Cyber to Daybreak Red customers on 5 November and its notice to Daybreak Blue users that it will train on their cyber sessions. Independent evaluations of the two new products have not yet been published.
