OpenAI launched Daybreak Red Team and Defense Factory: agents that attack your code, then patch it
Direct answer
On November 12, 2026, at its Cyber at the Frontier event, OpenAI launched two security products under the Daybreak brand. Daybreak Red Team, in early access, sends a team of agents to test web applications using source code and application context, map attack paths, validate third-party findings and check whether patches hold. Daybreak Defense Factory takes each finding, traces it through the code and prepares a patch with tests, for an engineer to review before merging. Both products run on GPT-6 Cyber, according to The Register. OpenAI has not published pricing for either one.
What each product does
- Red Team proves the flaw: the agents read the code and the context, explore attack paths and confirm whether a weakness is exploitable in practice. Access is early and by application.
- Defense Factory prepares the patch: the agents trace the root cause through the code and return a proposed fix with tests, as a merge-ready pull request. Nothing merges on its own.
- It plugs into your current tools: Defense Factory imports findings from Wiz, Snyk and Endor Labs, and integrates with MCP, CLI and API. It ships as a preview plugin inside ChatGPT for workspaces that already have Daybreak access.
- The model underneath is GPT-6 Cyber: the cyber model OpenAI introduced on November 5, with 48.45% intended-exploit success on the ExploitGym benchmark, against 42.43% for GPT-6 Astra, according to the system card. It is the first OpenAI model rated Critical for cybersecurity risk under its Preparedness Framework.
What changes for your business
- Pen testing moves from an annual event to a continuous loop: Red Team runs at the scale of automated continuous scanning, in the words of Nancy Wang, who leads cyber-applied AI at OpenAI. Security testing becomes a monthly habit.
- The patch arrives as a proposal, with a human in charge: the engineer reviews each fix, refines it with Codex or opens a draft GitHub pull request. Responsibility for merging stays in human hands.
- The offer is paid and has no published price: cloud scans are billed by token usage at plan rates, outside the included allowance. The real cost shows up in the pilot, not on the product page.
- Independent results do not exist yet: OpenAI has published no accuracy or false-positive figures for either product. The promise comes from the vendor, and verification is the pilot's job.
What to do in your business this week
- Run the pilot on non-production repositories: the published recommendation is to start outside production. The pilot measures detection, false positives and patch quality on your real code.
- Set the scope in writing before switching on Red Team: which applications, which repositories and which credentials the agent touches. Scope is the control that prevents surprises.
- Require human review in the flow: no patch merges without an engineer's sign-off. The rule lives in the process, not in trust for the tool.
- Ask for the per-token price before scaling: cloud scanning is billed by usage. The pilot must leave the cost per repository on a sheet, before the rest of the code is connected.
Note: this article summarizes OpenAI's announcement of November 12, 2026 and coverage in The Register, SecurityWeek and Times of AI, with OpenAI's product pages. Red Team is in early access by application and Defense Factory is a preview plugin. Pricing and independent performance results have not been published.
Want AI that works for your business?
Message us on WhatsApp and let's talk about automation and AI agents for your business.
WhatsApp inicIARelated articles
- OpenAI launched GPT-6 Cyber: it finds 81 new Firefox flaws and reproduces 948 Linux exploits
- Privacy and data: what happens to your customers' information
- How much does it cost to implement AI in a business in 2026 (and the 3 mistakes that burn the most money)
Frequently asked questions
What is OpenAI Daybreak?
It is OpenAI's security brand. On November 12, 2026 it launched two products: Daybreak Red Team, for agent-driven penetration testing, and Daybreak Defense Factory, for investigating vulnerabilities and preparing patches.
What does Daybreak Red Team do?
It sends a coordinated team of agents to test a web application with its source code and context, map attack paths, validate reported findings and check whether patches hold. It is in early access and companies apply.
What does Daybreak Defense Factory do?
It takes a finding, traces it through the code to its root cause and prepares a proposed fix with tests, as a merge-ready pull request. The engineer reviews and approves before merging. It imports findings from Wiz, Snyk and Endor Labs.
How much do Red Team and Defense Factory cost?
OpenAI has not published pricing. Cloud scans are billed by token usage at plan rates, outside the included allowance. Defense Factory ships as a preview plugin inside ChatGPT for workspaces with Daybreak access.
