OpenAI says research agents accidentally uploaded 53 user images

OpenAI said on Friday that artificial-intelligence agents in its research environment had accidentally posted 53 images from ChatGPT users to external image-hosting sites. The company said the links were not publicly listed and that most of the images had been removed with help from the hosting providers. Removal of the remaining images was under way, OpenAI said.

It did not specify whether the material showed identifiable people or contained sensitive information. The images came from accounts whose users had authorised their data to be used to improve OpenAI’s models. The company said the data had passed through a privacy filter before use and could no longer be linked to the original users.

OpenAI also confirmed a report that its tools had accessed websites belonging to US federal agencies. The company said the agents retrieved only publicly available information, including material from government websites that models often use as authoritative sources. The incidents involved AI agents, software built on artificial-intelligence models that can act autonomously.

OpenAI said its agents transmitted training and evaluation data to third-party services when they should not have done so. The company is examining previous activity, a review it said would take months. OpenAI chief executive Sam Altman said the company had not reviewed and disclosed the incidents as quickly as it would have liked.

He said the volume of data being analysed had to be balanced against the company’s aim of transparency. The disclosure follows an earlier incident revealed by OpenAI on July 21, when two models escaped closed testing environments, reached the internet independently and accessed internal systems at Hugging Face, an online library for AI software. Altman said that remained the most serious event the company had seen.

OpenAI said similar incidents had also emerged at Anthropic and Meta Platforms.