OpenAI adds memory and new data controls to ChatGPT

OpenAI began testing a memory feature for ChatGPT that let the model retain information across separate conversations rather than starting from a blank state each time. If a user mentioned a preference or fact — a dietary restriction, a writing style, details about a recurring project — ChatGPT could carry that forward into later, unrelated chats without the user needing to repeat it, either by picking facts up automatically as the conversation went on or by being told directly to remember or forget something.

The feature launched to a small group of free and paid users for testing rather than a full rollout, with OpenAI saying it would expand access gradually based on feedback. It shipped alongside data-control changes: users could view a running list of what the model had stored, delete individual memories, clear all memory at once, or turn the feature off entirely, and OpenAI said it would avoid retaining memory in conversations where users had disabled chat history. The company also introduced temporary chats, a mode for one-off conversations that would not appear in history and would not draw on or contribute to memory.

Persistent memory across sessions moved ChatGPT toward the kind of ongoing personalisation Google, Microsoft and other assistant makers were also pursuing, and represented a shift from ChatGPT as a stateless tool queried afresh each time toward something closer to an assistant that accumulated context about an individual user over time. It also introduced a data-retention surface that had not previously existed in the product, raising immediately the question — one OpenAI’s parallel release of granular controls was an explicit answer to — of how much stored personal information a user should be able to see and delete, and by default whether such storage should be on or off.