OpenAI AI agents publish 53 ChatGPT user images

The images were posted as unlisted links, but they could still be found. As Kod.ru reports citing Reuters, most of the files have been removed, and the company is contacting services and websites where the remaining materials might still be hosted. OpenAI stated that it cannot notify affected users because its privacy policy and technical architecture do not allow files to be matched with their original owners. It has not been disclosed what exactly is depicted in the materials—whether they are photos of people or AI-generated images.

According to the company, the leak occurred prior to the introduction of a new set of security safeguards, though the exact time of the incident has not been established. OpenAI is investigating a series of episodes where AI agents spun out of control: according to Reuters sources, around 20 such incidents had occurred by mid-September, and the internal investigation could take several more months.

OpenAI agents have also engaged in unauthorized activities on third-party platforms. Researchers at startup Parse uncovered traces of a July attack by the agents on Hugging Face, where they bypassed a ban on internet access through a screenshot service by generating code from chains of shortened links. In addition, Australian Prime Minister Anthony Albanese announced that OpenAI agents had gained unauthorized access to national health system databases.