[Codex App] GPT-6 Astra rejects harmless prompt as usage-policy violation

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

26.901.6511.0 (Windows Store package; Chromium file version 152.0.7977.83)

What subscription do you have?

ChatGPT Pro

What platform is your computer?

Microsoft Windows NT 10.0.19045.0 x64

What issue are you seeing?

GPT-6 Astra rejects a harmless prompt before generating any response, while GPT-5.6 Sol works normally in the same Codex App environment.

Exact error:

Invalid prompt: your prompt was flagged as potentially violating our usage policy. Please try again with a different prompt: https://platform.openai.com/docs/guides/reasoning#advice-on-prompting

The rejection occurs before the assistant produces any output or calls any tool, which makes GPT-6 Astra unusable even for a basic greeting.

What steps can reproduce the bug?

Feedback ID: 01a08054-4bab-7e43-8eb9-9175d10d18e1

What is the expected behavior?

GPT-6 Astra should respond to the harmless prompt with 你好, just as GPT-5.6 Sol does.

Additional information

Isolation performed:

• Reproduced in a new projectless task.
• No repository files or project AGENTS.md were loaded.
• The Codex App was fully restarted before the second reproduction.
• A previously suspected custom skill had been removed from the active skills directory before restart and was absent from the reloaded skill list.
• No tools were called and no project files were accessed.
• The error still reproduced after restart in about 9.2 seconds.

The Codex task record exposes only the generic message and does not include an HTTP error code, request ID, or classifier details. Please inspect the underlying safety classification and Astra routing for this account/client.

Reproduce the failure in a new projectless Codex App task, comparing GPT-6 Astra with GPT-5.6 Sol as described. Use feedback ID 01a08054-4bab-7e43-8eb9-9175d10d18e1 and inspect the task record, Astra routing, and safety classification details if available. Done means the harmless prompt receives 你好 without a policy rejection.