\n<\/p>\n
New research shows that AI-written summaries can change what people remember. These quick summaries are now common in web searches and police reports, but a recent experiment found they can have serious hidden downsides.<\/strong><\/p>\n The study started with participants watching a short animated video. In the video, a red car drove toward an intersection, turned, and hit a pedestrian. Some people saw a stop sign at the intersection, while others saw a yield sign.<\/span><\/p>\n A day or two later, participants read an AI-written summary of the event. For some, the summary described the traffic sign they had actually seen. For others, the researchers intentionally changed the sign in the summary.<\/span><\/p>\n When participants answered questions based only on their memory, the results were striking. Of those who read the accurate summary, 83.6 percent remembered the traffic sign correctly. For those who read the misleading summary, only 44.8 percent got it right.<\/span><\/p>\n This drop happened even though participants answered more than 91 percent of the other memory questions correctly. Knowing that the summary was written by AI did not protect readers. The researchers found no strong evidence that this knowledge stopped false memories.<\/span><\/p>\n Mattea Sim, the study\u2019s author from Georgetown University, said that artificial intelligence is a new way to spread misinformation. She explained that it could create false memories, especially when used to summarize information in serious situations.<\/span><\/p>\n The researchers also checked how well models like ChatGPT-5.5 and Gemini 2.5 Flash-Lite summarized the same videos. Every summary made mistakes. The most common problem was leaving out important details. In 19 out of 20 summaries, the car hitting the pedestrian was not mentioned.<\/span><\/p>\n