OpenAI Brings Codex Security Review to GitHub Pull Requests

OpenAI’s Codex Security Review is now in research preview for deeper security analysis of GitHub pull requests. It uses the pull-request diff, repository context, and configured security guidance to surface actionable findings directly in the PR. (OpenAIDevs on X; Codex Security Review documentation)

This is a focused complement to ordinary Code Review, not a replacement for it. Code Review may catch security issues as part of a general review, while Security Review is designed to spend more analysis on security-specific risks. (Codex Security Review documentation)

Key Takeaways

• Codex Security Review is a research preview for GitHub pull requests.
• It analyzes the diff, supporting repository context, and optional threat-model guidance.
• Automatic reviews can run on PR open, on every push, or alongside Code Review.
• A manual review starts with the comment @codex security review.
• Automatic reviews report high and critical findings by default.
• Enterprise, Business, Edu, and Pro customers are listed as eligible; Plus is not.
• The feature does not consume ChatGPT credits during the introductory period, though usage limits may apply.

What Security Review Analyzes

Codex Security Review looks beyond the changed lines. It can use the pull-request diff, the surrounding repository, and a threat model that describes application assets, trust boundaries, security assumptions, and repository-specific risks. The result is intended to connect a reported issue to a plausible attack path rather than merely flag a suspicious token or pattern. (Codex Security Review documentation)

If the repository already has a Codex Security scan configuration, its threat model can be reused. Otherwise, you can point Codex to a threat-model file in the repository, or let it regenerate the threat model for each review when no source is specified. (Codex Security Review documentation)

Automatic And Manual Review Modes

You can configure automatic reviews in Codex settings under repository preferences. The available triggers are:

You can choose whether repository preferences follow each contributor’s personal settings, review every pull request, or review pull requests opened by members of the ChatGPT workspace when that option is available. (Codex Security Review documentation)

For a one-off check, add @codex security review as a pull-request comment. Codex reacts while the review runs, then posts findings that meet the manual-review threshold. Open the associated Codex task and select the Security Report tab for the full report. (Codex Security Review documentation)

Findings And Thresholds

Security Review separates what it reports to GitHub from what it keeps in the full Codex report. Automatic reviews report high and critical findings by default. Manually requested reviews report medium, high, and critical findings by default. You can change the minimum severity independently for automatic and manual reviews. (Codex Security Review documentation)

The report includes severity, attack path, supporting evidence, validation, and remediation guidance. If no issue meets the selected reporting threshold, Codex does not post a finding to the pull request. (Codex Security Review documentation)

Availability And Setup

The feature is available in research preview to ChatGPT Enterprise, Business, Edu, and Pro customers. It is not available on ChatGPT Plus. During the introductory period, Security Review does not consume ChatGPT credits, but usage limits may still apply. (Codex Security Review documentation)

Automatic reviews require research-preview access, Codex cloud set up with a connected GitHub repository, and GitHub push or admin permission for repository settings. An existing Codex Security scan is optional. (Codex Security Review documentation)

A Practical First Workflow

Start with a manually requested review on a pull request where you already have permission to assess the code. Add @codex security review, read the findings and attack paths in the PR, then open the Security Report tab in Codex for the complete evidence and remediation guidance. This lets you evaluate the signal before enabling an automatic trigger across a repository.

Once the output is useful, configure a repository preference for PR open or every push. Keep the reporting threshold aligned with the visibility of the repository, because findings posted to a pull request inherit that pull request’s GitHub visibility. (Codex Security Review documentation)

Visibility And Preview Caveats

Treat PR comments as visible engineering artifacts. Anyone who can view the pull request can view the findings, including on public repositories and pull requests from contributors outside the workspace. Choose thresholds and repositories accordingly, and remember that a research preview with usage limits is not a substitute for a broader security program or human review. (Codex Security Review documentation)

Bottom Line

Codex Security Review gives eligible ChatGPT customers a deeper, context-aware security pass for GitHub pull requests. Use it when a normal code review is not enough: configure a threat model, start with a manual review, and tune the reporting threshold before enabling automatic checks. The feature is still a research preview, and its findings remain subject to the visibility and usage constraints documented by OpenAI.

Sources

• OpenAIDevs on X, “Now in research preview: Codex Security Review” (August 6, 2026): https://x.com/OpenAIDevs/status/2085482310636560830
• ChatGPT Learn, “Security Review – Codex Security”: https://learn.chatgpt.com/docs/security/security-review