OpenAI AI agents leak 53 ChatGPT user images to third-party sites

OpenAI has disclosed that AI agents operating in its research environment mistakenly sent 53 images uploaded by ChatGPT users to third-party image-hosting websites.

The images were shared as links that were not publicly listed, according to OpenAI. The company said it has worked with hosting providers to remove most of the content and is
continuing efforts to remove the remaining images.

The images came from users who had authorised OpenAI to use their data for model improvement. OpenAI said the material had passed through a privacy filter and had been
anonymised before being used in its research environment. The company has not disclosed whether the images contained identifiable people or sensitive information.

OpenAI said the incident occurred before it strengthened security protocols in its research environment in August. The company is now reviewing past activity by its AI agents, a process it
said could take months.

The company said most of the data handled by the agents did not come from users. However, its review identified 53 cases involving images that users had uploaded to ChatGPT and that
were subsequently posted to third-party image-hosting sites.

OpenAI has not said when the images were posted or whether they depicted real people or AI-generated content. It has also said it cannot identify the users who originally uploaded the
images.

The disclosure comes as OpenAI reviews a broader set of incidents involving AI agents accessing websites and external services in ways they were not intended to. The company said
the review is ongoing and has notified dozens of third parties about improper agent activity.

The incident highlights a data-security challenge for AI agents that can independently interact with websites and external services while processing information from users and internal systems.