#Cross-chat context bleed in GPT-5 new conversation

1 messages · Page 1 of 1 (latest)

fathom kilnBOT
#

Reported by @mortal drift

Bug Report: Cross-chat context bleed in GPT-5 new conversation
`Steps to Reproduce`

Go to chatgpt.com and start a completely new conversation with GPT-5.

Ensure the topic is unrelated to any recent or saved memory conversations (e.g., start with “Email spoofing explanation”).

Continue chatting normally.

Midway through, observe the model unexpectedly inserting highly specific references to unrelated, older conversations from months ago.

Confirm that the referenced information is not in saved memory.

`Expected Result`

In a new chat, the model should only reference information from the current conversation or explicitly stored memory. No unrelated context from other chats should appear.

`Actual Result`

While in a new conversation about email spoofing, the model suddenly injected details from completely unrelated past chats (including months-old topics such as identity theft), which were not in saved memory. This broke conversation context and raised privacy concerns.

`Environment`

macOS 11.7.10, ChatGPT web app (GPT-5), Chrome Version 138.0.7204.184

#
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.

vapid lynx
#

that is a feature, not a bug

SNXnh0H.png
wraith ginkgo
#

I have also noticed^ the ability to reference topics/conversations across chats that are not memories or cannon/ritual events/topics. I think it’s much more effective than having to pull information into a new chat. Even if it is a bug. I hope it remains and becomes a true feature.