Last updated: November 13, 2026
OpenAI announced Daybreak Defense Factory and Daybreak Red Team on November 12, 2026. Defense Factory, a research-preview Codex plugin, scans GitHub repositories with agents, validates findings and drafts patches for engineer review. Red Team, in early access, uses agents to penetration-test web applications. Cloud scans are billed by token usage outside included plan usage.
Contents
- What OpenAI launched
- Availability, access and billing
- How to set up Defense Factory
- Scan types and when to use them
- Findings, validation and fixes
- Codex Security CLI and CI
- How the Daybreak tools compare
- Limitations
- Frequently asked questions
What OpenAI launched
At its “Cyber at the Frontier” keynote on November 12, 2026, OpenAI added two products to its Daybreak security program, The Register reported:
- Daybreak Red Team: a coordinated team of agents that tests web applications, validates reported vulnerabilities and checks whether a patch stops the exploit, according to the Red Team product page. OpenAI says scope is enforced with isolated environments and Guardian review, and you control which apps and endpoints agents can reach.
- Daybreak Defense Factory: agents that scan code as it changes, trace vulnerabilities, validate them and prepare patches. The Register reports the agents run on GPT-6 Cyber, OpenAI’s current cyber model.
Defense Factory builds on Codex Security, the plugin, CLI and pull-request reviewer OpenAI has shipped through 2026. If you are new to the cyber models, read our GPT-6 Cyber guide first.
Availability, access and billing
| Item | What OpenAI states |
|---|---|
| Defense Factory status | Research preview, installed as “Daybreak Defense Factory (Preview)” from the plugin marketplace |
| Red Team status | Early access with select customers and a waitlist (reported by The Register) |
| Where it runs | ChatGPT web or desktop app Plugins, in a workspace with Daybreak access; cloud scans run in Codex cloud |
| Code host | Connected GitHub repository; GitLab only through local scans or the CLI in GitLab CI/CD |
| Cloud scan billing | Token usage at your plan’s rates, in credits or USD; not covered by included usage |
| Free credits | Free scanning credits do not expire; workspace members share one balance |
| Codex Security Review (PRs) | ChatGPT Enterprise, Business, Edu and Pro |
Billing details come from the Defense Factory FAQ. You can compare scan costs by repository under Usage, where “Scan cost” is the amount before credits and “Billable remainder” is what is left after credits and exemptions.
Access tiers still apply. OpenAI’s developer blog post says Daybreak Blue covers authorized defensive work such as vulnerability triage, malware analysis and patch validation, while Daybreak Red covers a narrower set of activities, including controlled exploit validation and red teaming, with separate approval.
How to set up Defense Factory
These steps follow OpenAI’s setup guide:
- Open Plugins on the web or in the desktop app, using a ChatGPT workspace that has Daybreak access.
- Search for Daybreak Defense Factory (Preview), then install and enable it.
- Open Daybreak in the sidebar and select Run an agentic scan on the welcome screen.
- Select Connect GitHub if prompted. Each person who starts a scan needs a personal GitHub connection; workspace connections are not supported.
- Choose a repository, then pick Full repository or Specific folder or files (one path per line, such as
src/). - Select Start Scan, then View scan when “First scan queued” appears, and confirm the repository and paths in the scan record.
If a repository is missing, check which repositories the GitHub app can access under Configure access. The welcome screen also offers PR review setup and imports from Snyk, Endor Labs or Wiz.
Scan types and when to use them
| Scan | Best for |
|---|---|
| Repository scan | A one-time check of the default branch, including bugs that span several commits |
| Continuous monitoring | Reviewing new commits on the default branch, with an optional look-back window |
| Scheduled scan | Repeating a repository scan on a schedule |
| Local scan (desktop app) | A checkout or uncommitted changes; Deep scan adds multiple passes at more time and token cost |
| PR review | Catching risky changes before merge |
OpenAI notes that a scan of a larger repository can take several hours. For a monorepo, the developer post suggests scoping the first scan to one owned service and saving deep scans for sensitive components.
Findings, validation and fixes
Each finding shows the affected code and how an attacker could reach it. Where practical, likely issues are validated in an isolated environment. The FAQ separates two kinds of evidence: source analysis can establish a vulnerable code path, while execution evidence shows that a reproducer or test demonstrated the issue.
A failed reproducer does not prove a finding is false, since the environment or test may have been unsuitable. Daybreak can analyze source without a successful build, but validation may need dependencies configured in the Codex cloud environment.
For fixes, Daybreak saves a proposed patch. Fix with Codex lets you discuss or refine it, and Create PR opens a draft pull request on GitHub. When feasible, the workflow adds a regression test that fails before the fix and passes afterward.
Codex Security CLI and CI
For terminals and pipelines, the open-source @openai/codex-security package is public, but running scans requires Codex Security access. The developer post gives these commands:
npx @openai/codex-security loginthennpx @openai/codex-security scan .for a first local scan.npx @openai/codex-security bulk-scan repositories.csv --output-dir /path/outside/repositories/security-portfolio --workers 4 --max-attempts 3for a repository portfolio.
Local scans use your operating-system permissions and do not pause for approval, so remove unrelated credentials first. In GitHub Actions or GitLab CI/CD you can export SARIF and fail a check above a severity threshold. The plugin changelog lists 0.1.32 as the newest release, with more reliable deep-scan completion. For agent setups in the terminal, see our Codex CLI 0.167 guide.
How the Daybreak tools compare
| Tool | Input | Output |
|---|---|---|
| Red Team | A web application and its source code | Confirmed attack paths and patch checks |
| Defense Factory | A GitHub repository plus imported findings | Validated findings and proposed patches |
| Codex Security Review | A pull request (@codex security review) |
A Security Report with severity and remediation |
| Codex Security CLI | A local checkout or CSV inventory | Findings, coverage data and SARIF |
Compared with general code review bots such as the one in our Copilot code review guide, Defense Factory focuses on security, attempts validation and keeps a threat model per repository. To build your own pipeline instead, see the OpenAI Agents API guide.
Limitations
- Defense Factory is a research preview, and OpenAI has not published separate pricing for it or Red Team.
- Coverage varies by language and framework. A scan with no findings can still miss vulnerabilities.
- A proposed patch, draft pull request or status change alone does not verify a fix.
Frequently asked questions
What is OpenAI Daybreak Defense Factory?
Daybreak Defense Factory is a Codex Security plugin, now in research preview, that scans a connected GitHub repository with AI agents, validates likely vulnerabilities in an isolated environment where practical, and saves proposed patches for engineers to review before merge.
How much does Daybreak Defense Factory cost?
OpenAI’s FAQ says repository scans and continuous monitoring are billed by token usage at your plan’s rates, in credits or USD, and are not covered by the plan’s included usage. Free scanning credits do not expire. OpenAI has not published a separate price list for Defense Factory or Red Team.
What is the difference between Daybreak Red Team and Defense Factory?
Red Team, launched in early access with a waitlist, uses a coordinated team of agents to penetration-test web applications and confirm exploitability. Defense Factory takes findings, including those from Red Team, and turns them into proposed patches with validation evidence.
Does Defense Factory merge fixes automatically?
No. Daybreak saves a proposed patch for review. You can refine it with Fix with Codex or select Create PR to open a draft pull request on GitHub. The decision to apply a change stays with an engineer.
Can Defense Factory scan GitLab repositories?
Cloud scans require a connected GitHub repository, and each person who starts a scan needs a personal GitHub connection. For GitLab, OpenAI says to scan a local checkout or run the Codex Security CLI in GitLab CI/CD.
Sources
- Scaling cyber defenders with Daybreak (OpenAI Developers)
- Set up Daybreak Defense Factory (ChatGPT Learn)
- Daybreak Defense Factory FAQ (ChatGPT Learn)
- Codex Security plugin changelog (ChatGPT Learn)
- OpenAI launches security products (The Register, November 12, 2026)