An MIT Media Lab experiment found that people shown AI-altered images and videos were more likely to misremember the originals, and feel more confident in those mistakes. The strongest effect came from AI-generated videos based on edited photos, which produced 2.05 times as many false memories as the control group.
Watch Desk analysis
What happened
The experiment used 24 ordinary photographs and 200 participants, divided into four groups of 50. According to PetaPixel’s account of reporting by The Washington Post, participants saw either the original photographs, AI-edited versions, videos generated from the original photographs, or videos generated from AI-edited versions. They were later asked questions about the original images.
All three groups exposed to AI-generated or AI-edited material produced more false memories than the control group. The largest effect was in the group shown videos generated from edited photos. That group also reported confidence in incorrect answers at 1.19 times the control group’s level. The research was published at the CHI 2025 conference, PetaPixel says. Read PetaPixel’s account of the study.
Why it matters
The finding is not simply that synthetic images can fool the eye. In this experiment, exposure was associated with mistakes about what people had seen before, and greater confidence in some of those mistakes. That matters wherever edited images or generated video are used as evidence, records or reminders.
The result also gives a concrete reason to be careful about treating vivid media as a reliable memory aid. It does not show that every AI-edited image will alter someone’s memory, or establish how long the effect lasts outside this experiment. But “I’m sure I remember it” may be a less sturdy test than it sounds when the images have been changed.
Our read
This is a meaningful warning about a subtle consequence of generative AI: synthetic media may affect not just what people see, but how confidently they recall it. The most useful response is not to distrust every photograph, but to preserve originals and check provenance when an image matters. A convincing clip is not a time machine, however much it would like the job.
What to watch
- Whether further studies reproduce the effect with different participants, images and real-world viewing conditions.
- Whether researchers test how long false memories and confidence effects persist.
- Whether platforms and newsrooms make original files and edit histories easier to identify.
Discussion spark: Should newsrooms and social platforms be expected to preserve and label original images, or is it ultimately the viewer’s responsibility to check what they are seeing?
Sources and evidence
- Source update (19 October 2026, 12:41 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.