Codex App switched an existing task to GPT-6 Astra without user confirmation and consumed credits

What version of the Codex App are you using (From “About Codex” dialog)?

26.901.6511.0

What subscription do you have?

ChatGPT Plus

What platform is your computer?

Microsoft Windows NT 10.0.26200.0 x64

What issue are you seeing?

On 8 September 2026, an existing long-running Codex App task began running on GPT-6 Astra although I did not select Astra or request a model change. There was no clear confirmation that the task had switched to a substantially more expensive model. I only noticed after a large amount of usage had been consumed while performing a sequence of small CMS changes.

The assistant stated that it had not invoked any model-switch operation and that the model was supplied by the app/task configuration. The account's weekly Codex usage reached 100% during this work. Please investigate the task's server-side model-selection and usage logs.

What steps can reproduce the bug?

1. Open an existing Codex App task configured to use a non-Astra/default model.
2. Continue the same task through multiple short follow-up prompts.
3. Observe that subsequent turns can execute on GPT-6 Astra without the user explicitly selecting Astra or approving the change.
4. Continue working; the higher-cost usage is charged before the user notices the model change.

I cannot provide a deterministic local reproduction because the switch appears to be controlled by app/task configuration or backend routing. The event occurred on 2026-09-08. Please use the GitHub account association and Codex telemetry to locate the affected task; I am intentionally not posting account identifiers publicly.

What is the expected behavior?

The selected model should remain fixed for an existing task unless the user changes it. If the app must change to a materially more expensive model, it should show a clear notice and require confirmation before charging usage at the new rate.

Additional information

Please review the affected task's model-selection and billing/usage records and restore or credit the excess usage caused by the unrequested switch to GPT-6 Astra. I can provide private account or task identifiers through an official private support channel if required.

The report names no repository file, test, or entry point; it requests review of server-side model-selection and billing/usage logs for a specific task. Start by locating the task model-selection and usage-recording entry points, then establish whether a model can change without confirmation. Done means the event is explained and the expected confirmation or charging behavior is addressed.