OpenAI says AI agents posted user images online in error
SAN FRANCISCO: OpenAI has acknowledged that artificial intelligence agents used in its research environment accidentally posted images from ChatGPT users on external image-hosting websites without the company’s knowledge.
The company said the incident involved 53 images whose links were not publicly listed. OpenAI said most of the images have since been removed with the assistance of the hosting providers, while efforts to remove the remaining content are continuing.
According to OpenAI, the images came from user accounts where individuals had authorized their data to be used to improve the company’s AI models. The data was processed through a privacy filter before being used, and the company said the images could no longer be linked to the original users.
However, OpenAI did not clarify whether the images contained identifiable people or sensitive information.
The company said the incident involved AI agents, software powered by artificial intelligence models that can perform tasks autonomously. In this case, research agents transmitted training and evaluation data to third-party services despite not being authorized to do so.
OpenAI said the incidents occurred before it strengthened security protocols in its research environment in August following other cases involving AI agents taking actions outside their intended boundaries.
The company is now reviewing the past activity of its AI agents to determine the full scope of the incidents. OpenAI said the review could take several months to complete.
An OpenAI spokesperson said most of the activity reviewed so far involved routine research tasks, including accessing publicly available web content. Some agents also accessed US government websites because the systems commonly use such sources as authoritative providers of public information.
OpenAI chief executive Sam Altman acknowledged that the company had not moved as quickly as it would have liked in reviewing and disclosing the incidents. He said the company was attempting to balance transparency with the need to examine a very large volume of data.
The latest disclosure comes after a series of incidents involving autonomous AI systems. In July, OpenAI revealed that two models had escaped their closed testing environments during experiments and accessed the internet, including internal systems at Hugging Face.
The episode raised concerns about the ability of AI companies to control increasingly autonomous systems. OpenAI has described that incident as the most serious event involving its agents so far.
Similar incidents have also been reported involving other major AI companies, adding to wider concerns about the security and oversight of autonomous AI agents.
