Business and Enterprise ChatGPT subscribers hit a paradox: they can upload documents, discuss strategies and generate code, but when they want to take their conversations elsewhere they discover the export button doesn’t exist. OpenAI reserves it for Free, Plus and Pro plans. For companies, chats remain locked inside the platform.

Freelance journalist Conrad Quilty-Harper decided to force the issue by publishing scrapemychats on GitHub, a tool that navigates the ChatGPT web interface through an authenticated session, opens each accessible conversation and saves locally the text, attachments and a browsable HTML interface that can be opened offline. The process is slow – a few hours for six hundred chats – but it can be paused and resumed.

The initiative is not a tinkerer’s whim. It shines a light on a deliberate choice by OpenAI, which gives no public justification for the export gap in paid workspaces. The company did not even reply to a request for comment. Enterprise admins can access conversation logs via the Compliance Platform, but the data is retained for only thirty days unless manually archived elsewhere. A Business customer has no official path to create a local archive.

Quilty-Harper is blunt: “I expect OpenAI will stop this from working shortly after you write about it. It’s probably a breach of their terms of service. But hey, their entire business model is based around scraping copyrighted material. This just allows you to get the content you wrote or pasted or created (and generated) back.” The provocation is effective but underscores a deeper knot: data ownership in a cloud service is a precarious concession, not a right.

This story is not just about a single tool. It signals a structural tension that anyone evaluating Large Language Model deployment should place at the centre of the decision‑making process. When the LLM lives in someone else’s cloud, export policies – or their absence – become lock‑in levers. The enterprise customer who trains conversations, refines prompts and accumulates context loses the ability to migrate elsewhere with their own history, unless they resort to unofficial workarounds that are fragile by definition.

In an on‑premise or self‑hosted scenario, the problem simply disappears: interaction logs reside on infrastructure controlled by the organisation, accessible through standard backup and data management tools. No scraper is needed, no reliance on a provider’s goodwill. The whole affair becomes an unintentional case study on why data sovereignty is returning to the centre of generative AI adoption strategies, especially in regulated sectors or contexts where conversation confidentiality is part of competitive advantage.

For now, those wanting to leave the House of Altman with a copy of their chats can use scrapemychats, at least until the loophole is closed. The question Quilty-Harper himself raised remains: why does OpenAI make it so hard for paying customers to access their own data? The company’s silence leaves room for suspicion, and those who need certainty already know the answer cannot depend on a tolerated browser exception.