›_ Codex Issues
Codex Issues / #32224

Auto code reviews continue with Auto review disabled and consume main model usage

Open 💬 2 comments Opened Jul 10, 2026 by rldyourmnd

Summary

Codex continues to run automatic GitHub code reviews even when Auto review is disabled. These reviews consume the account's normal model usage.

This is not a separate Code Review quota issue. The unexpected reviews consume the main ChatGPT/Codex model usage that is needed for regular work.

Account and setup

  • ChatGPT Pro
  • Codex in ChatGPT Cloud
  • GitHub account: rldyourmnd
  • GitHub organization: NDDev-it-com
  • GitHub Connector is needed for normal ChatGPT/Codex work with repositories, including repositories owned by the organization.

Expected behavior

When:

  • Personal Auto review is disabled; and
  • repository-level Auto code review is set to Follow personal preferences;

Codex should not automatically start code reviews for pull requests. It should only review a pull request when explicitly requested.

The GitHub Connector should remain usable for normal repository-related ChatGPT/Codex tasks.

Actual behavior

Automatic code reviews still start despite the settings above and consume normal model usage.

The Code Review settings page showed:

  • Personal Auto review: disabled
  • Exhaustive code review: disabled
  • Repository Auto code review: Follow personal preferences
  • Repository review trigger: Follow personal preferences

Despite this, new/recent pull requests appeared in the Codex Code reviews view and Codex performed review work.

Reproduction

  1. Connect GitHub to ChatGPT/Codex.
  2. Keep the GitHub Connector enabled because repository access is needed for normal ChatGPT/Codex conversations.
  3. Set personal Auto review to disabled.
  4. Set repository Auto code review to Follow personal preferences.
  5. Open or update a pull request in a connected repository.
  6. Observe that Codex still starts a review.
  7. Observe that the review consumes the main model usage.

Additional investigation

The account had both a personal GitHub App installation and an organization-level ChatGPT Codex Connector installation. The organization installation was temporarily suspended during investigation, then restored because suspending it removed the repository access needed by ChatGPT/Codex.

The following did not provide a reliable way to keep repository access while stopping the reviews:

  • disabling personal Auto review;
  • disabling exhaustive review;
  • setting all visible repositories to Follow personal preferences;
  • disconnecting and reconnecting the ChatGPT GitHub Connector.

The only reliable stop was removing/revoking the GitHub connector, but that also removed the required repository access.

Impact

  • Unexpected autonomous review tasks run against pull requests.
  • Main model usage is consumed without an explicit request.
  • Users must choose between keeping GitHub repository access and preventing unwanted reviews.
  • The UI does not expose which installation, account, workspace, repository rule, or webhook caused a review.

Requested fixes

  1. Make the disabled Auto review state authoritative across all personal and repository-level settings.
  2. Ensure Follow personal preferences resolves to disabled when personal Auto review is disabled.
  3. Keep normal GitHub repository access independent from automatic code review.
  4. Show the effective review source for every repository:
  • personal preference;
  • repository override;
  • organization/workspace policy;
  • explicit @codex request.
  1. Show the GitHub installation and webhook event that triggered each review.
  2. Add a clear account-wide “Disable all automatic GitHub reviews” control.
  3. Ensure automatic reviews cannot consume main model usage when the user has disabled them.

Related reports:

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗