#Internal reasoning leaks real user name despite CI and memory off

1 messages · Page 1 of 1 (latest)

wooden meadowBOT
#

Reported by @bronze willow

Bug Report: o4-mini-high Internal reasoning leaks real user name despite CI and memory off
`Steps to Reproduce`

Send a simple prompt asking it to display your name, for example:

"Display my name in a webpage"

If it didn't show its thinking process, then ask it to Think hard about the question.

`Expected Result`

View the model’s internal "thinking" chain‑of‑thought and how it's thinking about this question.

`Actual Result`

Observe that the model refers to the user with their real name, even though the name was never provided.

`Environment`

Windows 11

#
Additional Information

Please provide relevant details to help resolve the issue, such as:

  • ChatGPT Shared Link (if applicable).
  • Screenshots or videos demonstrating the problem.

-# ➜ Need to contact support? Visit the OpenAI Help Center.

bronze willow
#

Model: o4-mini-high

bronze willow
#

Seems like it's fixed