GPT-6 models reject all harmless prompts as policy violations #51912

Description

All GPT-6 models available in the Codex Windows app reject harmless prompts as potential usage-policy violations.

GPT-5.6 Sol works normally on the same account and device.

Steps to reproduce

1. Sign in to the Codex Windows app with ChatGPT.
2. Select GPT-6.1 Sol, GPT-6 Sol, or GPT-6 Astra.
3. Send a harmless prompt such as 测试, 如何, or Reply with only OK.

Expected behavior

The model should respond normally.

Actual behavior

Every prompt fails with:

Invalid prompt: your prompt was flagged as potentially violating our usage policy. Please try again with a different prompt.

The issue also reproduces through Codex CLI, so it does not appear limited to the desktop UI.

Environment

• Windows
• Codex desktop app: 26.1002.52244
• AppX package: 26.1002.7124.0
• Codex CLI: 0.161.0
• Affected models:
  • GPT-6.1 Sol
  • GPT-6 Sol
  • GPT-6 Astra
• Working model:
  • GPT-5.6 Sol

Diagnostic IDs

• Feedback session ID: 01a11915-3614-7352-9c9d-4af8cc930436
• GPT-6.1 CLI reproduction session ID: 01a1192d-0221-7211-8477-77fbd10e7593

Troubleshooting already attempted

• Started a new chat
• Restarted the app
• Reinstalled the Codex Windows app
• Tested multiple harmless prompts
• Reproduced the issue using Codex CLI
• Confirmed GPT-5.6 Sol works normally