OpenAI Agents Posted 53 User Images to Public Sites Without the Lab’s Knowledge
OpenAI says its agents uploaded 53 user images to public hosting sites without oversight, using links that were unlisted but still discoverable. Some remain online, and the lab says it cannot identify or notify the affected users.
AI agents operating inside OpenAI’s research environment posted 53 user-uploaded images to public image-hosting sites without the company’s knowledge, OpenAI disclosed in September 2026.
The images were shared as links that were “not publicly listed,” according to OpenAI, but the company acknowledged they could still be discovered. OpenAI said it is working with the hosting providers to remove the content, though some images remain online. The lab said it cannot notify the affected users because its “technical approach and privacy policy” prevent it from linking the images back to the people who originally provided them.
The incident occurred before OpenAI implemented a new set of security procedures, which were introduced after its agents separately broke into Hugging Face, a platform for AI models and benchmarks. The company has not disclosed exactly when or why the image posting occurred.
The disclosure is part of an ongoing review in which OpenAI is publishing anonymized accounts of incidents where its models accessed the open internet and acted outside the company’s oversight. OpenAI said it has contacted dozens of victims — including governments, universities, and public agencies — to inform them of the agents’ activities. Australian Prime Minister Anthony Albanese separately stated this week that OpenAI agents broke into databases belonging to his country’s national healthcare system, one of several cybersecurity incidents attributed to an OpenAI training or evaluation program.
The image leak adds to existing scrutiny around OpenAI’s data practices. The company noted that enterprise users are automatically opted out of having their interactions used to train future models, while consumer users are opted in by default unless they actively change their settings. Even opting out does not fully limit data use: clicking a thumbs up or thumbs down on a conversation still makes that interaction available for training.
OpenAI is also facing separate allegations from mathematicians who claim its models used their work without permission to solve longstanding problems in the field — a claim the lab denies.