OpenAI Daybreak Red Team and Defense Factory: AI pentest agents

OpenAI now sells AI agents that attack your web app and agents that write the patch. At its Cyber at the Frontier keynote on November 12, OpenAI launched Daybreak Red Team, an AI penetration testing product for web applications, and Daybreak Defense Factory, which turns security findings into proposed fixes. Neither has a public price. The useful part for a small team is the design, and it costs nothing to copy.

What actually happened

From OpenAI's own product pages:

• Red Team puts "a coordinated team of agents" on your web apps. They use source code and app context to investigate attack paths and confirm vulnerabilities. They also check findings from code scanners, bug bounty reports and disclosure reports, then test whether a patch actually stops the exploit.
• Scope controls. OpenAI says Red Team runs in isolated environments with "Guardian review," and you control which apps and endpoints the agents can reach.
• Defense Factory takes findings from your existing security tools and from Red Team. Agents trace each issue through your code and develop a patch. You can work on the fix with Codex, scan new commits, or schedule recurring checks.
• Humans merge. "Engineers review and test proposed patches before merge," per OpenAI. The Register reports the agents currently run on GPT-6 Cyber.
• Access and price. Red Team is "register interest." Defense Factory is "contact sales." The Deep View reports OpenAI gave no pricing information.

Greg Brockman told the keynote audience that companies will be secure only to the extent of the frontier model they use to defend themselves, The Register reports. That's a sales line from the company selling the model. Treat it like one.

Why it matters for your business

Most small companies don't get breached through a clever zero-day. They get breached through the known issue Dependabot flagged eight months ago that nobody had time to fix.

Defense Factory targets that gap. You don't need OpenAI's sales team to copy the workflow:

1. Collect the findings you already have. Dependabot, npm audit, GitHub secret scanning, your WAF logs. Put them in one queue.
2. Let an agent draft the fix. Claude Code, Codex or Copilot can open a pull request with the patch and a test.
3. A human merges. Always. CI runs the tests. A person reads the diff. The agent never pushes to main.
4. Scope the attacker agents hard. If you ever run AI pentesting, point it at staging, list the endpoints it can reach, and log every request.

Finding bugs is getting cheap. Fixing and reviewing them is the bottleneck now.

Sources: OpenAI Daybreak Red Team, OpenAI Daybreak Defense Factory, The Register, The Deep View.