One wrong detail in an AI summary can alter a person's memory

Misleading AI summaries can reshape what people remember, according to new research presented by a team led by Mattea Sim from Georgetown University and co-authored with researchers at the University of Washington. The study finds that when a summary includes one incorrect detail, participants become far less able to recall the original event accurately. The work will be presented at the Ninth AAAI/ACM Conference on AI, Ethics and Society in October 2026. In tests, AI summaries frequently made errors, leaving out an average of 51.6% of central details, including 95% that skipped a key moment in a video. In a memory experiment with 331 participants, accurate summaries produced 83.6% correct recall, versus 44.8% after misleading summaries.