ChatGPT self-generates 96% of user memories, study finds

Ninety-six percent of the memory entries ChatGPT held about a sample of real users were generated by the chatbot itself rather than requested, according to a study presented at the ACM Web Conference 2026 and reported by Futurism. Researchers led by Abhisek Dash pulled 2,050 memory entries from 65 of 80 users whose chat logs were obtained via European data privacy requests.

About 28 percent of those entries contained personal data as defined by Europe's General Data Protection Regulation, and 52 percent contained what the paper calls 'psychological insights' about users. In one example the researchers cite, a user who asked for bands like Nirvana was logged as someone who 'likes Nirvana,' a determination the paper says 'is not explicitly stated and relies on an assumption.'

Dash framed the stake plainly. 'There is no boundary anymore; the system is acting as your advisor, your friend, your doctor, your lawyer, and so it has information on everything,' he said. He also pointed at retention design: 'If the system knows you so well, and if its objective is to cater to your beliefs and validate your thoughts sycophantically to have you come back to it time and again, then it can have disastrous broader implications.'

OpenAI allows users to opt out of memory and in January introduced what it described as 'a more capable and scalable system for synthesizing memory.' The company did not, in the reporting, dispute the study's figures.