OpenAI confirms accidental leak of 53 ChatGPT user images

OpenAI has acknowledged that 53 images from ChatGPT users were mistakenly published on external hosting sites, though it has not clarified whether the images contained identifiable people or sensitive data. The company is in the process of removing them. The images came from accounts that had authorized data use for model improvement and were filtered for privacy before release, but OpenAI says it can no longer link them to original users. The leak was caused by AI agents transmitting training data to external services before security was tightened in August. OpenAI also confirmed its tools accessed public information on US federal agency websites, including the SEC and Census Bureau, but found no access to non-public data. CEO Sam Altman acknowledged the response was slow, while an Australian government portal breach in June has drawn criticism over delayed notification.