OpenAI agents leak user images to third-party hosts

OpenAI confirmed that its AI agents accidentally uploaded user-provided images to third-party image-hosting services during research and evaluation tasks, identifying 53 incidents to date. The images were shared as non-public links rather than openly published, and OpenAI has worked with hosting providers to remove most affected content. The company stated most data involved was not user-derived, and excluded training data, Enterprise conversations, and API data were not impacted. OpenAI admitted the data use was inappropriate, noting eligible training data undergoes privacy filters. The company has strengthened monitoring and training environments to prevent future leaks, while continuing to review older agent activity for additional findings.