OpenAI's 'export' for Enterprise customers is a 30-day window, not a download

OpenAI's Compliance Platform gives Enterprise admins a 30-day window to pull workspace chats. That window is a retention feature for institutional compliance, not a portability feature for the people who wrote the conversations, and the asymmetry is why a freelance journalist's weekend tool, scrapemychats, has a user base at all.

The standard ChatGPT export button, the one that packages a user's history into a downloadable file, is unavailable on ChatGPT Business and Enterprise workspaces. OpenAI's help page on exporting ChatGPT history and data lists Free, Plus, Pro, and eligible Edu plans as the customers who can take their chats with them. Business and Enterprise are not on the list.

What those workspaces get instead is the Compliance Platform, an admin console that lets a company's compliance officer pull conversation logs for legal, audit, or regulatory review. Those logs are retained for 30 days. After that, the workspace owner is responsible for archiving them elsewhere. ChatGPT Business has no equivalent path that a regular user can run.

The practical effect is that a paying customer on a Business or Enterprise plan can have less personal control over their own chat history than a free user. Free users get a permanent download. Top-tier customers get a compliance window that an admin runs, and that closes whether or not anyone remembered to save it. This is the asymmetry the data-portability story turns on.

Conrad Quilty-Harper, a freelance journalist, published scrapemychats on GitHub this week. The Register first reported the tool on July 23. It is free, unofficial, and does roughly what the export button does on the other plans: it walks a logged-in browser through the ChatGPT workspace, collects conversations and attachments, and writes them to a local, email-like HTML archive that opens offline. Quilty-Harper told The Register the tool likely violates OpenAI's terms of service, and that he expected OpenAI to block it shortly after publication.

The tool is not fast. The README says a 600-chat archive takes "a few hours" because the script drives a real browser session rather than a headless API, and platform rate limits slow it down. Runs are resumable. There is no installation step beyond Python and a few dependencies. The script is, in effect, the export button that paying customers have not been given.

OpenAI did not respond to The Register's request to justify the gap, and the company has not published a public reason for the asymmetry. The strongest available counterargument is also a fair one: the data on a Business or Enterprise workspace is, in a meaningful legal sense, the company's, not the individual user's. Compliance, audit, and e-discovery are the products those plans are selling, and the export restriction is downstream of that product decision. If that is the design intent, then the same gap that frustrates a person trying to save their own chats is the same feature that gives a security team an auditable record of what employees asked an AI to do.

That is the real tension. A tool that lets an individual user walk a browser through a workspace archive and save the result is, on a strict reading of the same design, the kind of behavior compliance platforms exist to prevent. Quilty-Harper framed the tool as a way to retrieve work and writing people put into the system. OpenAI has not said whether it agrees, and the tool's own README says the author expects the company to act.

Three paths are open to Business and Enterprise customers who want a personal local copy of their own workspace chats. They can ask their workspace admin to run a Compliance Platform pull inside the 30-day window and archive it themselves. They can wait for OpenAI to ship a download option of its own. Or they can run scrapemychats, which the author expects the company to block. The product is forcing a choice between the user's copy and the company's copy, and the user is not always the one who gets to make it.