User reports GPT-6 still sees the old version of a message after you edit it
A forum post says GPT-6 in ChatGPT keeps pre-edit text in view. It is unverified, but you can test it in a few minutes.
The short version
- One forum post says GPT-6 on ChatGPT web still sees earlier versions of edited messages.
- The poster says GPT-5.6 Sol does not do this and only sees the current conversation.
- The poster tested on a Plus plan and is unsure whether the behavior is intentional.
- It is a single, unverified report. No official response has appeared in our sources.
Why it matters
People edit messages to remove a mistake, a draft figure or a name they should not have shared. If the model still reads the old wording, editing is not the same as deleting. Until this is confirmed or fixed, treat an edit as a cosmetic change, not a privacy tool.
What the user reported
A user post on community.openai.com says GPT-6 in the ChatGPT web app seems to keep every version of an edited message in its context. Context is the text the model reads before it writes each reply. The poster says GPT-5.6 Sol does not behave this way.
The poster also says they are not sure whether this is intentional, and that they saw it on a Plus plan.
How the poster tested it
The steps are simple. Start a chat with GPT-6 and send a throwaway greeting. Then send a second message labeled "Test 1" and edit it so it reads "Test 2". Finally, ask the model to list the earlier user messages it can see.
According to the post, GPT-6 still listed the "Test 1" wording even though that version no longer appeared on screen. When the poster switched the same chat to GPT-5.6 Sol and edited again, that model saw only the current conversation.
What we do not know
This is one post from one user. We have not seen OpenAI confirm it, explain it, or call it a bug, and we cannot say whether it happens on other plans or for other people.
It is also unclear how much the hidden text affects answers in everyday use. The test asks the model to recite old messages directly, which is a stricter check than normal chatting.
What to do in the meantime
If you rely on editing to fix a slip, be careful. The safe habit is to start a new chat when a message contains something sensitive, such as client names or private figures.
Running the test yourself takes a few minutes and tells you whether your own account behaves the same way.
What to watch
- Does OpenAI confirm the behavior and say whether it is intended?
- Do other users reproduce it, and on which plans?
Try it
In a fresh GPT-6 chat, state a made-up detail, edit that message to a different detail, then ask the model which details you gave. If the old one appears, start a new chat instead of editing.
Sources and links
First reported by community.openai.com. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.