
Daybreak Red Team and Defense Factory: How AI Pen Tests Work
OpenAI's Daybreak Red Team and Defense Factory pair AI pen testing with automated patching. Here's how the 2 new security agents find, fix and retest.
Finding a vulnerability has never been the hard part of application security. Proving it is real, ranking it against everything else in the queue, writing the fix, and then checking that the fix actually holds is where security teams lose weeks. OpenAI is now aiming its Daybreak security platform squarely at that loop. On November 12, the company introduced two new products, Daybreak Red Team and Daybreak Defense Factory, which use coordinated AI agents to attack an application, prepare patches, and retest them.
- Daybreak Red Team: early-access AI penetration testing for web applications, using source code and app context to confirm exploitable flaws and retest patches
- Daybreak Defense Factory: a research-preview plugin that connects to a GitHub repository, traces findings through the code and prepares proposed fixes for engineer review
- Guardrails: assessments run in isolated environments with defined boundaries, and Guardian review controls which applications and endpoints the agents can reach
- Launch date: announced November 12, 2026 at OpenAI's Cyber at the Frontier event; OpenAI has not published pricing
What Does Daybreak Red Team Actually Do?
Red Team is the offensive half. According to OpenAI's description, relayed by Help Net Security, a fleet of agents studies an application's source code and context, investigates possible attack paths, and confirms which vulnerabilities are genuinely exploitable. Nancy Wang, who leads OpenAI's cyber-applied AI work, described it to The Register as pen testing "at the scale of automated continuous scanning."
The most practical feature may be validation. Red Team can take findings that already exist, from code scanners, bug bounty submissions and vulnerability disclosure reports, and check whether each one is real. For a team drowning in unverified reports, that turns a pile of maybe-problems into a ranked list backed by evidence. Once a patch is written, Red Team can run the exploit again to confirm the fix blocks it.
How Defense Factory Turns Findings Into Patches
Defense Factory handles remediation. Its agents pick up findings from existing security tools or from Red Team, trace each vulnerability through the codebase, and prepare a proposed fix. Engineers stay in charge: patches are reviewed and tested by people before anything merges, and teams can use Codex to examine the evidence, add context and refine a change.
Setup is deliberately simple. Teams with access install the Daybreak Defense Factory (Preview) plugin in ChatGPT and connect a GitHub repository. Scans run in Codex cloud and can cover a repository's whole default branch or selected folders and files. Teams can also:
- Monitor new commits on the default branch
- Schedule recurring scans
- Run security reviews on pull requests before they merge
- Scan local checkouts and uncommitted changes from the desktop app, with an optional Deep scan mode that makes multiple passes
Why Pairing Offense and Defense Matters for AI Security
Security vendors have offered scanners and automated pen tests for years. What is new here is connecting both ends of the vulnerability management workflow in one agent system, so the evidence that proves a flaw exists flows straight into the fix, and the fix flows straight back into a retest. That closes the gap where most findings stall.
It also builds on a busy year for the platform. Daybreak earlier split into separate red and blue defender tiers, and OpenAI committed $1 billion to frontline cyber defenders. Red Team and Defense Factory are the first Daybreak pieces aimed squarely at everyday application security teams rather than specialist researchers.
What Should Security Teams Watch Before Adopting It?
A few practical questions remain open. OpenAI has not published pricing, Red Team requires an early-access application, and Defense Factory is labeled a preview. Each person starting a cloud scan needs a personal GitHub connection, which security leads will want to map against their access policies. And as with any AI-generated patch, the human review step is the safety net, so teams should keep it mandatory rather than optional.
The direction is still encouraging. AI-driven bug hunting is pushing disclosure volumes to record highs this year, and defenders need tools that verify and fix at the same pace. A system that confirms, patches and retests in one loop is exactly the kind of help stretched AppSec teams have been asking for. For more on hardening agent tooling, see our MCP server security checklist and the rest of our AI security coverage.
Sources: Help Net Security — November 13, 2026; The Register — November 12, 2026; ITdaily — November 13, 2026.
More Ai Security Stories

Apple Secure Exclaves: How Your Camera Data Stays Private
Apple's Secure Exclaves keep raw camera and mic data away from the main OS, apps and Apple itself. Here is how the 3-part design protects ambient sensing.

Project Zero Full 90 Policy: What Faster Disclosure Means
Google Project Zero's Full 90 policy drops the 30-day patch grace period and moves to 60 days in 2027 and 30 in 2028. Here's what defenders should do.

ChatGPT Wallet Security: How Agents Pay Without Your Card
ChatGPT Wallet lets AI agents buy things using one-off payment tokens and a Face ID check on every purchase. Here's how the security model protects you.
