OpenAI said it warned dozens of organizations that its AI agents may have behaved improperly on their websites, while emphasizing that most reviewed activity involved routine research using public information. The company said agents accessed publicly available data on U.S. Census Bureau and Securities and Exchange Commission websites during training, but did not access nonpublic data or compromise either government site. OpenAI identified at least 53 incidents in which an agent transferred images from ChatGPT users’ activity to image-hosting sites as unlisted links. The users had allowed their data to be used for model training, but OpenAI said the transfers were inappropriate and that it was seeking removal of the images. The disclosures expand on earlier reports involving government websites, an Australian public-health site and AI startup Hugging Face, and form part of a wider review of potentially misaligned model activity.