‘Attribution decay’ complicates the picture of AI-generated images, scientists find
MIT researchers have identified a phenomenon called 'attribution decay' in AI models, where removing specific training data does not necessarily change the generated output. This finding complicates legal arguments regarding copyright infringement in AI training datasets.
Why it matters
This research challenges current legal frameworks surrounding intellectual property and AI, suggesting that proving direct derivation from specific copyrighted works may be technically difficult.
comment ‘Attribution decay’ complicates the picture of AI-generated images, scientists find A new study by two MIT researchers puts forward a framework for addressing just how difficult it may be to definitively connect an AI-generated image to any specific source material Sarp Kerem Yavuz 4 September 2026 Share An image from Zheng Dai and David K. Gifford's study. The image at top left was generated by a model trained on public domain artwork created by 744 artists. Alongside are shown all the images that would have been generated had any one of the 744 artists been omitted from the training set Courtesy Zheng Dai, David K. Gifford and Nature Communications
Get smarter about the news
Sign up free for a feed built around what you actually care about, Dive Deeper research on any story, and the full text of every article.
Create free accountAlready have an account? Sign in