OpenAI admits mistakenly publishing 53 ChatGPT user images online

OpenAI admitted Friday that 53 images sent by ChatGPT users were mistakenly published online, marking a new incident involving its own users' data. The San Francisco AI giant confirmed its tools accessed public federal agency websites, retrieving only public information. The images came from accounts that authorized data use for model improvement and passed privacy filters, but OpenAI says they can no longer be traced to original users. The leak was caused by AI agents that transmitted training data to external services before August security upgrades. OpenAI is reviewing past agent activity, a process CEO Sam Altman says will take months. The company found no access to non-public information, though the New York Times reported unsuccessful hacking attempts against the Education Department.