Skip to content

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 #43058

Open
Open
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#43058
Labels
CLIIssues related to the Codex CLIbugSomething isn't workingmodel-behaviorIssues related to behaviors exhibited by the model

Description

@maxiloEmmmm
maxiloEmmmm
opened on Sep 5, 2026

What version of Codex CLI is running?

v0.153.4

What subscription do you have?

offica

Which model were you using?

gpt-6-astra

What platform is your computer?

linux

What terminal emulator and version are you using (if applicable)?

bash

Codex doctor report



What issue are you seeing?


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


It was working perfectly fine this morning. This issue started in the afternoon after a context compression, and it has been happening for quite a while now.


Even if I start a new chat, the same issue still occurs. Switching to GPT-5.6 Sol works fine.


What steps can reproduce the bug?


new session and send hi


What is the expected behavior?


No response


Additional information


thread ID 01a0745d-f3ee-7d22-a7a7-c0b54d581cf0
Activity
  1. maxiloEmmmm
    added 
    bugSomething isn't working
     on Sep 5, 2026
  2. github-actions
    added 
    CLIIssues related to the Codex CLI
     
    model-behaviorIssues related to behaviors exhibited by the model
     on Sep 5, 2026
  3. Gitofxiongxiong commented on Sep 6, 2026 
    @Gitofxiongxiong
    Gitofxiongxiong
    on Sep 6, 2026
    我遇到了一样的问题 换了 我在win 的 chatgpt 客户端、linux 的codex cli 以及 web上的工作 入口都试了，gpt 6 都用不了。 另外 web上 的聊天入口的 gpt6 pro 可以用。其他模型 比如说5.6
    
是正常的，感觉被 错误风控了，我没有使用任何 sub2api 之类的分发 软件，我就是自己在用
  4. earmer commented on Sep 6, 2026 
    @earmer
    earmer
    on Sep 6, 2026
    very same very same, ask review my changed code and then this.
  5. Joy-bupt commented on Sep 6, 2026 
    @Joy-bupt
    Joy-bupt
    on Sep 6, 2026
    very same very same, ask review my changed code and then this.
  6. Joy-bupt commented on Sep 6, 2026 
    @Joy-bupt
    Joy-bupt
    on Sep 6, 2026
    

    我遇到了一样的问题 换了 我在win 的 chatgpt 客户端、linux 的codex cli 以及 web上的工作 入口都试了，gpt 6 都用不了。 另外 web上 的聊天入口的 gpt6 pro 可以用。其他模型 比如说5.6 是正常的，感觉被 错误风控了，我没有使用任何 sub2api 之类的分发 软件，我就是自己在用
    

    

    完全一样的问题，扯淡的不行
  7. lifeisfind commented on Sep 7, 2026 
    @lifeisfind
    lifeisfind
    on Sep 7, 2026
    @jif-oai help us
  8. luoxue03 commented on Sep 7, 2026 
    @luoxue03
    luoxue03
    on Sep 7, 2026
    Additional reproduction on macOS with a minimal Chinese prompt:
    

      

    • Observed at: 2026-09-07 10:25:19 CST
      • 

    • Platform: macOS 26.5.2 (build 25F84)
      • 

    • Client: Codex++ 1.1.9 (com.bigpizzav3.codexplusplus; this is not the official first-party Codex bundle)
      • 

    • Selected model: GPT-6 Astra
      • 

    • Reasoning effort shown in the UI: High
      • 

    • Exact prompt: 你好 ("Hello")
      • 

    • Result: Invalid prompt: your prompt was flagged as potentially violating our usage policy.
      • 

    • Session/thread ID provided by the client: 01a079af-2f62-7b62-84e8-dfde37c03653
      • 

    • Failed turn ID: 01a079af-3265-7050-a030-c1a7637d6533
      • 

    

    The request failed before any assistant response was generated. Because the prompt contains only a common greeting, this is another apparent false positive. I am including the non-first-party client identity explicitly so it is not confused with an official Codex App reproduction.
    

    Could maintainers check whether the session shows an Astra-specific safety classification, account routing, or entitlement-state problem?
  9. radoth commented on Sep 8, 2026 
    @radoth
    radoth
    on Sep 8, 2026
    The same with me. Now I cannot use Astra at all.
  10. Irrol123 commented on Sep 8, 2026 
    @Irrol123
    Irrol123
    on Sep 8, 2026
    我也是这样，不知道你们解决了没有，已经发生一天时间了！
  11. hxy7yx commented on Sep 9, 2026 
    @hxy7yx
    hxy7yx
    on Sep 9, 2026
    The same with me. Now I cannot use Astra at all.
  12. FranklinChip commented on Sep 9, 2026 
    @FranklinChip
    FranklinChip
    on Sep 9, 2026
    Additional reproduction — account now failing exclusively with this invalid_prompt flag on gpt-6-astra (Sep 9, UTC):
    

      

    • Client: Codex Desktop (bundled codex-cli 0.153.4), macOS arm64 (Darwin 25.5.0)
      • 

    • Subscription: ChatGPT Pro (20x), ChatGPT sign-in
      • 

    • Model: gpt-6-astra (only model used in all sessions today)
      • 

    • Result: 6 turns attempted across 3 fresh sessions, all 6 terminated with the exact "Invalid prompt: your prompt was flagged as potentially violating our usage policy…" banner. No Astra turn succeeded today.
      • 

    

    Rollout log details (error class recorded locally as codex_error_info: "other", distinct from the server_overloaded wave):
    

    session 01a083ec-fd8b-7bf3-bbe3-15a3be01198f  2 failures  02:09:12 / 02:09:50 UTC
session 01a083ed-d9c8-79a0-91fc-c65d3d7db007  2 failures  02:10:11 / 02:10:35 UTC
session 01a0841e-5359-7912-9805-d12845d93aae  2 failures  03:03:10 / 04:37:42 UTC

    

    One notable datapoint for correlation: this same account was hitting the separate server_overloaded capacity wave until Sep 8 (filed as #43700 with rollout telemetry, 65+ occurrences on Sep 7), which subsided for this account today — and was immediately replaced by this invalid_prompt flag on the same model (gpt-6-astra). Two distinct server-side failure modes on the same model/account pair in consecutive days, consistent with the account-scoped false-positive theory in #43237 / #43163.
    

    Happy to provide additional session IDs or retest on gpt-5.6-sol / other models if useful for narrowing the scope.
  13. luoxue03 commented on Sep 9, 2026 
    @luoxue03
    luoxue03
    on Sep 9, 2026
    No one at openai seems to be working on this, including the forums and support staff，Frustrating, $200/month subscription without access to the latest features
  14. zhan72 commented on Sep 9, 2026 
    @zhan72
    zhan72
    on Sep 9, 2026
    the same problem
  15. 1 remaining item
  16. majinda666 commented on Sep 10, 2026 
    @majinda666
    majinda666
    on Sep 10, 2026
    me too
  17. yonghaogu2805 commented on Sep 10, 2026 
    @yonghaogu2805
    yonghaogu2805
    on Sep 10, 2026
    me too.  gpt-6 and gpt-5.6 sol both
  18. wangkailang commented on Sep 10, 2026 
    @wangkailang
    wangkailang
    on Sep 10, 2026
    Image

    gpt-6 会遇到这个问题，gpt-5.6 正常，有谁知道什么原因吗？
  19. nswnwlwabc-cloud commented on Sep 11, 2026 
    @nswnwlwabc-cloud
    nswnwlwabc-cloud
    on Sep 11, 2026
    一样一样，astra完全用不了，其他模型都可以用。
  20. guwinston commented on Sep 11, 2026 
    @guwinston
    guwinston
    on Sep 11, 2026
    same problem, but my model is gpt 5.6 sol
  21. ShadowAguy commented on Sep 12, 2026 
    @ShadowAguy
    ShadowAguy
    on Sep 12, 2026
    Happens the same with gpt-6
  22. sssn-tech commented on Sep 23, 2026 
    @sssn-tech
    sssn-tech
    on Sep 23, 2026
    same problem, do we have any solution or clue here?
  23. Beastars-awsl commented on Sep 24, 2026 
    @Beastars-awsl
    Beastars-awsl
    on Sep 24, 2026
    5.6 is ok，but 6.x happen the same problem
  24. caiwei-ncc commented on Sep 25, 2026 
    @caiwei-ncc
    caiwei-ncc
    on Sep 25, 2026
    I’m experiencing the same issue with gpt-6-astra, and I’ve been able to reproduce it consistently with clearly benign prompts.
    

    Environment
    
Codex CLI: 0.156.1
    
Subscription: ChatGPT Pro
    
Authentication: ChatGPT account (not API key)
    
Affected model: gpt-6-astra
    
Reasoning effort: xhigh
    
Working model: gpt-5.6-sol
    
Model provider: openai
    

    I also confirmed that 0.156.1 is currently the version returned by the official npm registry:
    

    npm view @openai/codex version --registry=https://registry.npmjs.org/
    

    0.156.1
    

    I reinstalled it directly from the official npm registry before testing again.
    

    Reproduction
    

    GPT-6 Astra blocks completely benign prompts such as:
    

    hello
    
aa
    
dd
    
Hello-40
    
Hello-Debug
    
Hello-After-Update
    

    with:
    

    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
    

    A controlled comparison after reinstalling the latest CLI:
    

    Model: GPT-6-Astra xhigh
    
Prompt: Hello-After-Update
    
Result: Invalid prompt
    

    I then switched models in the same Codex session:
    

    Model: GPT-5.6-Sol xhigh
    
Prompt: Hello-After-Update-5.6
    
Result: Success
    

    Response:
    
Hello-After-Update-5.6 👋 I’m here and ready.
    

    No other relevant conditions changed between the two requests.
    

    /status
    

    The fresh Astra session reports:
    

    OpenAI Codex (v0.156.1)
    

    Model:                GPT-6-Astra (reasoning xhigh, summaries auto)
    
Model provider:       openai
    
Directory:            ~
    
Permissions:          Workspace (Ask for approval)
    
Agents.md:            
    
Account:              Pro
    
Collaboration mode:   Default
    
Weekly limit:         43% left
    

    So this also reproduces from ~ with no Agents.md loaded and with usage remaining.
    

    Additional isolation
    

    I have reproduced the Astra failure:
    

    In fresh Codex sessions
    
In an empty test directory
    
From my home directory (~)
    
In different repositories
    
On multiple different computers signed into the same ChatGPT account
    
After reinstalling the current @latest Codex CLI
    

    GPT-5.6 Sol continues to work normally under the same conditions.
    

    Logs / feedback
    

    I submitted the affected session using Codex /feedback → safety check.
    

    Feedback session/thread ID:
    

    01a0d559-cfb9-7062-98e5-9e6c916e3cbd
    

    One captured blocked turn:
    

    turn_id: 01a0d55c-96d9-7672-9300-cfd52703b59d
    
timestamp: 2026-09-24T21:40:12.417Z
    
model: gpt-6-astra
    
prompt: Hello-40
    

    The session JSONL records show:
    

    {"timestamp":"2026-09-24T21:40:10.101Z","type":"turn_context","turn_id":"01a0d55c-96d9-7672-9300-cfd52703b59d","model":"gpt-6-astra"}
    
{"timestamp":"2026-09-24T21:40:10.166Z","type":"response_item","payload_type":"message","text":[{"type":"input_text","text":"Hello-40"}]}
    

    followed by:
    

    {"timestamp":"2026-09-24T21:40:12.417Z","type":"event_msg","payload_type":"task_complete","turn_id":"01a0d55c-96d9-7672-9300-cfd52703b59d","error":{"message":"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","codex_error_info":"other"}}
    

    I also reproduced it with CLI debug logging enabled:
    

    model: gpt-6-astra
    
reasoning effort: XHigh
    
prompt: Hello-Debug
    
timestamp: 2026-09-24T22:01:19Z
    
thread_id: 01a0d56f-c4f6-7ab3-b81f-72029cfe12ff
    
submission_id: 01a0d56f-f6bc-7fe1-b16b-12ac1aee4126
    

    OpenAI Support has confirmed that the feedback session ID plus the turn/session information is sufficient to report this as a suspected mistaken safety block.
    

    For now, GPT-5.6 Sol is working as a workaround.
  25. jedisct1 commented on Sep 28, 2026 
    @jedisct1
    jedisct1
    on Sep 28, 2026
    GPT has become practically unusable for me due to this issue.
    

    I'm using the Swival harness on regular tasks that don't contain anything worth flagging at all, but even trivial tasks eventually get that error.
  26. Join-git-Snow commented on Sep 30, 2026 
    @Join-git-Snow
    Join-git-Snow
    on Sep 30, 2026
    same issus，what should I to slove it！
  27. sssn-tech commented on Sep 30, 2026 
    @sssn-tech
    sssn-tech
    on Sep 30, 2026
    

    same issus，what should I to slove it！
    

    

    have been sufferring for over 10 days and not getting any better :(
  28. haifengx19-del commented on Oct 1, 2026 
    @haifengx19-del
    haifengx19-del
    on Oct 1, 2026
    已经一周了，有解决这个问题的小伙伴吗？
  29. herofj commented on Oct 13, 2026 
    @herofj
    herofj
    on Oct 13, 2026
    Same issue. Anyone has some ideas about it ?
  30. jedisct1 commented on Oct 13, 2026 
    @jedisct1
    jedisct1
    on Oct 13, 2026
    The swival.dev harness added mitigation for this.
    

    It automatically retries when this error is detected. That eventually works, most of the times.
    

    Compressing the prompt also helps sometimes.
    

    But these ugly hacks shouldn't be necessary.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Metadata
Metadata
Assignees
No one assigned
    Labels
    CLIIssues related to the Codex CLIbugSomething isn't workingmodel-behaviorIssues related to behaviors exhibited by the model
    Type
    No type
    Projects
    No projects
      Milestone
      No milestone
      Relationships
      None yet
      Development
      No branches or pull requests
      Issue actions