OpenAI Launches AI Agents That Find and Patch Software Flaws

OpenAI introduced two security products on Thursday that let AI agents attack an application, confirm the flaws, and draft the fixes. The launch, announced at OpenAI's Cyber at the Frontier keynote, pushes the company deeper into enterprise cybersecurity — a market where security teams have long faced more vulnerability reports than people to chase them down.

The two products are Daybreak Red Team and Daybreak Defense Factory. Red Team runs penetration tests on web applications and checks whether reported flaws can actually be exploited. Defense Factory takes those findings, traces each one through the code, and prepares a patch for an engineer to approve. Johannes Landgraf, OpenAI's head of cyber, said Red Team "paves the path for us to dramatically expand the access to our frontier cyber models over time," as The Deep View reported.

OpenAI President Greg Brockman announced the pair on X on November 12: "Both make it much easier to put frontier cyber capabilities to work, helping defend companies in new ways."

What does Red Team actually do?

Red Team puts what OpenAI calls "a coordinated team of agents" on a single application. The agents use source code and application context to investigate attack paths and confirm whether vulnerabilities are real. According to The Deep View, the work is divided up: one agent coordinates the assessment, others investigate the code, and reviewer agents check the results.

Customers can also feed the system findings from other sources. The product accepts output from code scanners, bug bounty submissions and disclosure reports, and uses it to prioritize what to fix first. Once a patch exists, teams can rerun a test to confirm it actually stops the exploit.

The agents don't roam free. Testing happens in isolated environments, and Guardian review controls which applications and endpoints the agents can reach. That constraint matters: an automated penetration tester with broad access is itself a security risk.

The takeaway: Red Team is a gated, multi-agent pipeline that turns raw vulnerability reports into confirmed, exploitable-or-not findings.

How does Defense Factory turn findings into patches?

Defense Factory handles the repair half of the loop. It can pull in findings from Red Team or from other tools, including Wiz, Snyk and Endor Labs, SecurityWeek reported. It can also scan connected GitHub repositories, watch new commits, and run scheduled checks on its own.

Engineers stay in the decision seat. They can examine the evidence behind a finding, adjust a proposed patch using OpenAI's Codex coding agent, and open a draft pull request. Nothing ships on its own — engineers review and test proposed patches before merge, and SecurityWeek noted that patches are not deployed automatically.

That human-in-the-loop design is the key mechanism here. The product automates the slow, repetitive parts of vulnerability triage and patch drafting, but a person still signs off on every change to production code.

The takeaway: Defense Factory drafts fixes, but engineers decide what merges.

Who is OpenAI aiming at?

OpenAI framed the launch around speed. "Attackers are able to move faster. I think we see that right now almost on a daily basis," Landgraf said in the briefing. "Defenders also have to move faster at that."

Clint Gibler, who leads cybersecurity research at OpenAI, positioned the products as a supplement to human red teams rather than a replacement. "There's always been a dearth of security expertise in the world," he said. "Our ability to secure the world needs to scale."

Fouad Matin, the company's cyber preparedness and safety lead, said OpenAI aimed to make the tools "turnkey and easy to use" for organizations that "might not have the red team staff on hand." That's the market gap OpenAI is targeting: companies with more code and more vulnerability reports than security headcount can support.

The stakes are straightforward. If autonomous agents can reliably confirm exploits and draft patches, the bottleneck in software security shifts from scarce human expertise to compute and review capacity. If they can't, the products add another layer of automated output that stretched teams must verify by hand.

The takeaway: OpenAI is betting that agent labor, not more hiring, closes the security staffing gap.

Where does this fit in OpenAI's security push?

The launch builds on a fast run of security moves. OpenAI released GPT-6 Cyber last week. Earlier this month, it brought in Hacktron — the research firm that found a flaw in OpenAI's own sign-on system — to test how its products connect with third-party services.

In the keynote, Brockman argued the industry needs "to build software in a fundamentally more secure way." The Daybreak products are the commercial expression of that argument: sell the models' offensive and defensive cyber capabilities as packaged tooling rather than raw API access.

The timing also matters for the market. Cybersecurity vendors already sell vulnerability scanning and patch management as separate products; OpenAI is selling both halves of the workflow — testing and repair — as one agent-driven loop.

The takeaway: Daybreak turns OpenAI's cyber research into a commercial pipeline spanning detection through patching.

How can companies get access, and what does it cost?

Access is limited for now. Red Team is in early access, and companies can register interest through OpenAI's form — though OpenAI says registering does not guarantee a spot. Defense Factory is available as a preview plugin for workspaces that already have Daybreak access, SecurityWeek reported.

OpenAI has not said what either product will cost, or when Red Team will open to a wider group of customers. Pricing and general availability will likely decide whether these products reach the mid-market organizations Matin described — the ones without red teams on staff — or stay confined to early-adopter enterprises already inside the Daybreak ecosystem.

The takeaway: the technology is here, but OpenAI hasn't yet said who gets it or at what price.