By Hamza Faisal

OpenAI is moving its artificial intelligence from writing code to defending it. The company has launched two new security products under its Daybreak platform that use fleets of AI agents to probe web applications for weaknesses and then patch them, a push it presented on Thursday as the next front in enterprise cybersecurity.

The products, Daybreak Red Team and Daybreak Defense Factory, were unveiled at OpenAI's "Cyber at the Frontier" keynote on November 12. OpenAI President Greg Brockman told the audience that the industry has reached a point where a company's security depends on the frontier model it uses for defence, and on how well it uses it.

A pen test that never sleeps

Daybreak Red Team is an automated penetration-testing service. A coordinator agent runs each assessment, investigator agents dig into source code and application context to work out whether a flaw can actually be exploited, and reviewer agents check the results before anything is reported. The idea is to give security teams a prioritised list of real, exploitable weaknesses rather than a raw dump of possible issues.

Nancy Wang, who leads cyber-applied AI at OpenAI, described the service as superhuman penetration testing delivered at the scale of automated, continuous scanning. Red Team is launching in early access with a small group of customers, and a waitlist opened on Thursday. OpenAI has not disclosed pricing for either product.

From finding flaws to fixing them

Daybreak Defense Factory takes the other half of the job. It automates what OpenAI calls the defensive loop: finding vulnerabilities, validating them, and fixing them at scale. Agents built on GPT-6 Cyber, the company's security-focused model, scan codebases, identify flaws, and draft patches for engineers to review before anything is merged.

Defense Factory is available through the Daybreak plugin in Codex, OpenAI's coding agent, putting the patching workflow directly inside the tool developers already use to write software. The company stressed that human review remains part of the process: the agents propose fixes, but engineers decide what ships.

Security's new dividing line

The launch extends a Daybreak programme that has accelerated sharply in recent weeks. OpenAI released GPT-6 Cyber, its most powerful security model, on November 5, and days earlier made its Daybreak Red model available in Amazon's GovCloud for approved government and defence customers. Daybreak Blue, the companion service for verified defenders, validates vulnerabilities and fixes with fewer unnecessary refusals.

The timing underlines how central cyber defence has become to the AI labs' pitch. The same companies building ever more capable models are now selling the tools to defend against them, and OpenAI is betting that enterprises will pay for agents that can test and repair their software around the clock. Whether security teams trust an AI to decide which flaws matter, and which patches are safe, will be the real test of the new products.

Hamza Faisal is the founder and editor of Paktechstan.