›_ Codex Issues
Codex Issues / #5530

Codex made a very concerning "typo" re: hardcoded superadmin

Resolved 💬 2 comments Opened Oct 22, 2025 by mattmireles Closed Nov 11, 2025
💡 Likely answer: A maintainer (etraut-openai, contributor) responded on this thread — see the highlighted reply below.

What version of Codex is running?

codex-cli 0.47.0

What subscription do you have?

ChatGPT Pro

Which model were you using?

GPT-5-Codex

What platform is your computer?

codex-cli 0.47.0

What issue are you seeing?

I'm setting up a new application as the solo developer. I instructed Codex to hard code my email address as the super admin with access to the admin portal—essentially a God view of all users. Later, it spelled out my email address as "matt@gist.si" instead of "matt@gist.is" When I asked where it got that, it claimed it was just a typo, but it worries me!

Maybe I entered it wrong? Maybe i introduced the typo? Totally possible!

I've never seen Codex make a typo before. Given the high stakes involved in getting this right, I'm very worried that the wrong email is in fact hard-coded in the system as the only admin and that this is going to fuck me later.

What steps can reproduce the bug?

Uploaded thread: 019a0a64-c10b-7011-a99c-bfdfa5d4ff86

What is the expected behavior?

Codex either spells my email correctly or identifies that the misspelled email was saved in the system

Additional information

_No response_

View original on GitHub ↗

2 Comments

techjoec · 10 months ago

I have a project that is XXXX-browsertester. It constantly 'typos' browertester. I personally, and made it try as well, scoured my entire filesystem..No hits. I never typoed it that way in my session logs even. But it STILL does it even with explicit instructions not to. I chalk it up to training data taking priority..

etraut-openai contributor · 9 months ago

Thanks for the bug report. We've shared the feedback with the team members who are responsible for training our codex models. We're tracking this feedback and similar reports so we can make the model more robust in future iterations.

If you see other model behavior that you'd like to report, please use the recently-added /feedback command in the CLI. That will provide us with additional details including logs and conversation text.