OpenAI admits accidental leak of 53 ChatGPT user images

OpenAI admitted Friday that its artificial intelligence tools mistakenly published 53 images users had sent via ChatGPT, marking a new privacy incident involving its own customers' data. The images were posted to external hosting sites via links, though not publicly indexed. Most have been removed with hosts' help, and removal of the rest is ongoing. The images came from accounts that opted into data usage for model improvement, and OpenAI says it can no longer link them to original users. OpenAI attributes the leak to AI agents that transmitted training data to external platforms before August security upgrades. The company is reviewing past agent activity, a process expected to take months. Affected sites include government, university, and public agency platforms. CEO Sam Altman acknowledged the response was slower than desired, citing petabytes of data to analyze.