OpenAI Agents Now Pen-Test and Patch Web Apps

OpenAI on Thursday, Nov. 12, launched two cybersecurity products that put teams of AI agents to work attacking and repairing customers' software. Daybreak Red Team probes web applications for exploitable flaws, while Daybreak Defense Factory writes patches for engineers to review, the company said at its "Cyber at the Frontier" keynote, as reported by The Register and The Deep View.

Key takeaways

• Daybreak Red Team uses a fleet of agents to test web apps for vulnerabilities and re-test fixes; it is in early access with a waitlist.
• Daybreak Defense Factory scans code, proposes patches and hands engineers a pull request with a test; it is offered through the Daybreak plugin in Codex.
• OpenAI disclosed no pricing, and the launch materials reviewed for this story included no accuracy or false-positive figures.

How Daybreak Red Team works

Red Team runs a group of agents against a customer's web application. One agent coordinates the assessment, others investigate the source code and application context, and reviewer agents check the results, according to The Deep View's account of an OpenAI press briefing.

The tool can also check whether vulnerabilities reported elsewhere, such as by code scanners, bug bounty programs or disclosure reports, are actually exploitable, and then test whether a patch closes them. Nancy Wang, who leads cyber-applied AI at OpenAI, called it "superhuman pen-test at the scale of automated continuous scanning," The Register reported.

Red Team is launching in early access with select customers and a waitlist, The Deep View reported. Johannes Landgraf, OpenAI's head of cyber, said the product "paves the path for us to dramatically expand the access to our frontier cyber models over time."

What Defense Factory adds

Defense Factory takes over after a flaw is found. Its agents scan codebases, trace vulnerabilities and draft fixes, and engineers review and approve the changes before they reach production. "This way engineers get a merge-ready pull request with the patch and a test that shows that it works," Wang said.

The product builds on Codex Security, OpenAI's existing code-scanning tool. Its agents watch code as it changes, review pull requests and investigate alerts from other security tools, then notify the engineer who owns the code and propose a patch in Codex. The Register reported that the agents currently run on GPT-6 Cyber, OpenAI's newest gated security model.

Why OpenAI is pitching AI pen-testing now

OpenAI president Greg Brockman framed the launch around speed. "We're now at a point where companies will be secure only to the extent of the frontier model that they use to defend themselves and how well they utilize that," he said in the keynote, while adding that security fundamentals still matter.

Clint Gibler, who leads cybersecurity research at OpenAI, said the products are not meant to replace a human red team but to help defenders keep up with the volume of AI-generated code.

Context

The products extend Daybreak, OpenAI's program for giving vetted defenders access to its cyber models. They arrive a week after OpenAI released GPT-6 Cyber to select approved Daybreak Red customers on Nov. 5. AI-driven attack tooling is maturing on the other side as well; we have reported how attackers ran agents that rebuilt their malware until scanners stopped catching it.

The same day, four nonprofits including Public Citizen sued OpenAI, Anthropic, Google, Meta and SpaceX's AI business in San Francisco Superior Court, asking a judge to halt development of AI systems more capable than their most advanced existing models, The Register reported.

What's next

Key open questions are how quickly OpenAI widens Red Team access beyond its first customers, what the products will cost and how often their findings and patches hold up in real codebases.

FAQ

Can any company sign up for Daybreak Red Team?

Not yet. It is in early access with select customers, and others can join a waitlist.

Does Defense Factory change code on its own?

No. Its agents propose patches, and engineers review and approve them before they go into production.

References

• OpenAI launches security products, tells orgs they're doomed if they don't use them — The Register (Nov. 12, 2026)
• OpenAI sees a new enterprise frontier in cyber — The Deep View (Nov. 12, 2026)