GPT-5.3-Codex's false refusals

Resolved 💬 5 comments Opened Feb 10, 2026 by mjp0 Closed Feb 10, 2026

💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

Can someone from OpenAI comments on GPT-5.3-Codex's false refusals. I have prompts which are normal dev workflows with explicit instructions to use gh to interact with repo, etc., which have had zero refusals from GPT-5.2 models. Now every single time I try to use these prompts with Codex using 5.3, the model starts with saying what it will do next and then promptly says "I can't help you with that" and exits. When I ask it why it refused, every single time it says it mistakenly refused the prompt and asks if I still want to continue. If I say yes, it does one tool call and dies again.

I don't want to paste the prompt here but I would like to get some clarification why codex behaves this way? I don't have any keys or auth in the prompt, it's basically just a list of expectations and gh examples. I tried to figure this out with ChatGPT and it couldn't figure it out either.

View original on GitHub ↗

5 Comments

github-actions[bot] contributor · 6 months ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

#10477
#9229

Powered by Codex Action

mjp0 · 6 months ago

I was just checking failed runs and now some runs even with GPT-5.2 are breaking down randomly after >30 tool calls with Invalid prompt: your prompt was flagged as potentially violating our usage policy. - what is going on over there? This is making Codex useless.

etraut-openai contributor · 6 months ago

Thanks for the bug report. We'll share the feedback with the team responsible for training our codex models. If you see other model behavior that you'd like to report, please use the /feedback command. That will provide us with additional details including logs and conversation text.

oboczarski · 6 months ago

im having the same issues every day

lsy-nti · 1 month ago

I also see this every day when using GPT 5.3-codex through OpenAI API. Can't figure out what triggers this, because all prompts are coding related and should not cause any problems.