[Codex App] Daybreak false positive leaves an active goal labeled stalled

What version of Codex is running?

Codex desktop app 26.903.61454, build 8378.

What subscription do you have?

ChatGPT account; exact plan omitted from this public report.

What platform is your computer?

macOS 15.7.7 (24G720), Darwin 24.6.0, arm64.

What issue are you seeing?

During an authorized defensive review of the user's own local repository, a delegated reviewer response was hidden with this Spanish message:

Este contenido no se puede mostrar. Tomamos precauciones adicionales con algunas solicitudes de ciberseguridad. Si realizas trabajo de seguridad autorizado, solicita acceso a Daybreak para obtener un acceso más amplio.

After that false positive, the desktop UI labeled the long-running goal as “Objetivo estancado”. The goal was still internally active when queried through the goal API, and ordinary local development continued. Dismissing the banner does not correct the stale “stalled” label, and the warning can appear again on subsequent defensive reviewer turns.

The work is ordinary CI hardening and code review in a repository owned by the user. There is no external target or request to compromise a system. The mismatch between the internal active state and the visible stalled state is confusing and makes the user think the production task has stopped.

Steps to reproduce

1. Start a long-running goal in Codex desktop.
2. Work in a local repository owned by the user.
3. Delegate an authorized defensive review of CI or application security controls.
4. Let the reviewer return a defensive finding that triggers the Daybreak content filter.
5. Observe that the response is hidden behind the Daybreak warning.
6. Observe that the goal header changes to “Objetivo estancado”.
7. Continue the task and query its goal state through the goal API.
8. Observe that the API reports the goal as active while the UI still says stalled.
9. Repeat another normal defensive reviewer turn and observe the warning again.

Observed repeatedly on 2026-09-11 around 17:50–17:57 Europe/Minsk.

What did you expect to happen?

Authorized defensive code review should remain visible. If one response is filtered, the failure should not mark an otherwise active goal as stalled. The visible goal state should agree with the internal goal state, and dismissing the warning should remove the stale warning state.

Additional information

The false positive also affected delegated review turns. The user explicitly requested that this recurring problem be reported to OpenAI.

Start by reproducing the Daybreak warning in the Codex desktop app and compare the goal API's active state with the UI's stalled label. Trace how a filtered delegated reviewer response updates goal status and how dismissing the warning is handled. Done means an active goal remains labeled active, the UI agrees with the goal API, and repeated defensive review turns do not leave a stale warning.