Researchers have found that exposure to AI-generated or AI-altered images can lead people to form false memories of events that never occurred. In one experiment, participants who viewed manipulated photographs showed a large increase in false recollections and expressed higher confidence in those inaccurate memories than those who saw the originals.
The findings come from work led by Pattie Maes, a professor at the MIT Media Lab. Her team showed 24 ordinary photographs (a wedding, a city street, a race) to 200 people. After a break, some participants saw the same photos again, while others viewed AI-altered versions with changed facial expressions, inserted military personnel or added signs of climate change. When the memory test came, Maes winced at the results. “I think of all this as very dangerous, basically,” Maes told The Washington Post.
Related research points to similar risks with AI-generated text
A separate study by researchers at Georgetown University and the University of Washington looked at how AI-generated summaries affect eyewitness recall. The team recruited 331 participants to watch a short animated video of a car striking a pedestrian. A day or two later, they read either an accurate or a misleading AI-generated summary of the event. Those who read the misleading summary correctly recalled key details only 44.8 percent of the time, compared with 83.6 percent for those who read an accurate one. The paper, posted to arXiv on Sept. 23, 2026, also found that labeling summaries as AI-produced did not prevent the memory distortion.
Lead author Mattea Sim, an assistant research professor at the Massive Data Institute in Georgetown’s McCourt School of Public Policy, said in Neuroscience News, “AI is a new method of delivering misinformation, and it has the potential to create these false memories for people who are reading that information.”
Other studies described by the Post found that people questioned by a generative AI chatbot after watching a video of a crime developed three times as many false memories, and that participants who generated ideas with and without AI later had trouble remembering which ideas were their own. Gizmodo pointed to a study from last year that found both deepfakes and AI-generated texts planted false memories about made-up political scandals and even reduced voting intention.
Human memory has always been open to suggestion, and the Georgetown team wrote that AI “has the potential to generate misinformation, even absent any adversarial intent, which can meaningfully impact human memory.” Police are already experimenting with AI summaries of body-cam footage, the Post reported, and courts routinely depend on eyewitness recollections in cases ranging from car accidents to sexual assault.
Sim’s team plans to move next from animated clips to real police body-worn camera footage, according to Neuroscience News.