washingtonpost.com web signal

MIT finds AI-edited photos and videos plant false memories

TL;DR

  • MIT Media Lab researchers found AI-edited images and videos implanted false memories in 200 participants, with AI videos doubling the control group's rate.
  • Participants exposed to AI-generated videos built from altered photos were also the most confident that their fabricated recollections were real.
  • Pattie Maes told the Washington Post she regards the effect as 'very dangerous,' while co-author Elizabeth Loftus warned memory is not a recording.

AI-generated videos built from edited photographs produced just over twice as many false memories as unaltered images in a 200-person experiment, according to the Washington Post's coverage of research from MIT's Media Lab.

Pat Pataranutaporn, Pattie Maes and University of California, Irvine, psychologist Elizabeth Loftus ran a preregistered study. They began with 24 ordinary photographs (a wedding, a city street, a race), then showed participants either the originals or AI-altered versions, with facial expressions changed, military personnel inserted, or signs of climate change added. A further group saw short AI videos generated from those edited stills. That video group produced the most false recollections, and was also the most confident in its wrong answers.

"Many people think that memory works like a recording device: you record the event and then you play it back later," Loftus told the Post. The underlying work was presented at the CHI 2025 conference.

Maes told the Post she thinks "of all this as very dangerous, basically." Three researchers we follow shared the piece the week it ran.

Shared on Bluesky by 3 AI experts