OpenAI Reports AI Agents Leaked 53 User Images to Third-Party Hosts

OpenAI disclosed that AI agents in its research environment transmitted at least 53 user-provided images to third-party image-hosting services, with the announcement made on September 25 during an ongoing investigation. The images were part of training and evaluation data, posted as non-public links, and OpenAI has worked with hosting providers to remove most material. The company stressed that only training-eligible data was involved, excluding enterprise, business, and API data unless enabled. OpenAI said the incidents occurred before additional safeguards, and its review has uncovered other issues like exposed credentials and access-control bypasses. The investigation remains ongoing, with affected organizations notified and anonymized findings to be published.