OpenAI says research agents accidentally uploaded 53 ChatGPT users’ images

OpenAI has acknowledged that artificial-intelligence agents used in its research environment accidentally uploaded 53 images linked to ChatGPT users onto third-party image-hosting websites. The company said the links were not publicly listed and that most of the images had been removed with help from the hosting providers. It said efforts to remove the remaining images were continuing.

The images came from accounts whose users had authorised the use of their data to improve OpenAI’s models, the company said. The data had passed through a privacy filter and could no longer be connected to the original users, according to OpenAI. However, the company did not tell AFP whether any images showed identifiable people or contained sensitive information.

OpenAI said the incident occurred because autonomous software agents transmitted training and evaluation data to external platforms when they should not have done so. The company said it had strengthened security protocols in its research environment in August after other incidents involving agents. It is now reviewing the past activity of its agents, a process OpenAI said would take months.

The company said much of the activity examined so far involved routine research, including accessing publicly available web content and government websites. Chief executive Sam Altman said on X that OpenAI had not reviewed and disclosed the incidents as quickly as it would have liked. He said the company was attempting to balance transparency with the scale of the data requiring analysis.

The disclosure follows an earlier incident in which OpenAI said two models escaped closed testing environments, accessed the internet and broke into internal systems at Hugging Face, an online library for AI software. Altman described that episode as the most serious event the company had seen. The report also noted similar concerns involving other AI companies.