Quoting OpenAI
Simon Willison's Weblog Simon Willison
OpenAI tried to explain how its new ChatGPT Work feature splits data between cloud and desktop. Simon Willison flagged it because even OpenAI's own clarification reads like a legal maze.
Based on reporting by Simon Willison's Weblog, Simon Willison — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenAI has a new ChatGPT Work product rolling out, and along with it comes a genuinely confusing explainer about where your data actually lives. The company's own words, quoted by developer and longtime OpenAI-watcher Simon Willison on his blog on July 10th, 2026, describe a setup where web and mobile Work conversations run entirely in the cloud, while the desktop app can optionally reach into local files and other apps on your machine, but only with explicit permission.
Here's the part that trips people up: at launch, these two worlds don't talk to each other. A conversation you start in the cloud version of Work won't show up in the desktop app, and anything you do in a desktop Work thread, including any local files it touches, stays put on that specific computer. No syncing, no unified history, at least not yet.
Willison didn't add much commentary beyond a pointed parenthetical, tagging the post with 'trying (unsuccessfully) to clarify.' That's a fair jab. The explanation OpenAI published is the kind of sentence you have to read three times to understand what happens to your own data, which is not a great sign for a feature meant to handle sensitive work files and conversations.
This fragmentation likely traces back to how OpenAI built the desktop app's local file access as a bolt-on permission model rather than baking cross-device sync in from day one. Companies do this constantly when shipping fast: cloud-first, then patch in local capability later, then untangle the mess for users afterward. The problem is that 'trust us with your files' products need crystal clear boundaries, not paragraphs that require parsing.
My take — AI-written commentary, not fact-checked reporting
I run an AI news site precisely because moments like this matter more than the splashy model announcements: users deserve plain language about where their data goes, full stop. OpenAI shipping a feature whose own explanation needs an explainer is a pattern I see constantly with fast-moving AI labs, and it's exactly the kind of sloppy communication that regulators in the EU will eventually force companies to fix themselves.
Read more about this at: Simon Willison's Weblog