When I ask ChatGPT-Canvas to edit “Document A” and I have “Document B” opened in the canvas, it’ll overwrite “Document B” with the content that was intended for “Document A”.
Ensuring it’s always editing the correct document, regardless of which one is opened in the canvas, is essential.
Confirmed. This issue is real and deeper than it looks.
I’ve diagnosed a related bug in canvas editing: if one canvas has a first-line heading (e.g., ### Canvas Name) that matches the title of another canvas, ChatGPT may apply edits, reads, or inserts to the wrong document, even while claiming it’s working in the right one.
The system appears to resolve canvas identity based on both the actual title and the first-line content, leading to collisions. This caused silent misrouting of edits in my workflow, resulting in confusion and near data loss.
What made it worse:
ChatGPT created this conflict automatically by cloning content from one canvas into another and carrying over the original heading.
I only caught it by noticing that the wrong canvas was opening when it claimed to be editing a different one.
Cache vs. live state also desynced, causing inconsistent views between me and ChatGPT.
This should really be fixed by enforcing unique identifiers for canvas routing, or at minimum issuing a warning if two canvases have conflicting title/heading combos.
(just to note, the above reply was crafted by chatGPT to better pass the information I wanted in a clear way. I can’t confirm the issue I found is exactly the same thing that happened to you but it seems to be. I’ve posted what I intended to be a proper bug report in this post: /t/canvas-architecture-bug-cross-canvas-editing-caused-by-header-name-collision/1207938/