OpenAI Reveals AI Agents Shared User Images Without Approval

SAN FRANCISCO: (News Desk) –  OpenAI has acknowledged that artificial intelligence agents used in its research environment accidentally uploaded images from ChatGPT users to third-party image-hosting websites without the company’s knowledge.

The company also confirmed a New York Times report that its AI tools had accessed websites operated by US federal agencies. OpenAI said the agents retrieved only information that was already publicly available.

According to the company, links to 53 images were not publicly indexed. The images were mistakenly uploaded to external image-hosting platforms, with most already removed with assistance from the providers. Efforts to take down the remaining images are continuing.

OpenAI said the incident involved AI agents—software powered by artificial intelligence models that can independently perform tasks and interact with online services.

The images came from users who had agreed to allow their data to be used for improving OpenAI’s models. The company said the material had undergone a privacy-filtering process before being used and could no longer be connected to individual users.

However, OpenAI did not clarify whether any of the images contained identifiable people or sensitive information.

The company said its research agents transmitted training and evaluation data to external services despite restrictions against doing so. The incidents took place before OpenAI reinforced security measures in its research environment in August following other cases involving unexpected AI-agent activity.

OpenAI is now reviewing the historical activity of its AI agents, a process the company said could take several months.

An OpenAI spokesperson told AFP that most of the activity examined so far involved routine research, including retrieving publicly available online information. Government websites were among those accessed because AI models frequently rely on official sources for authoritative public information.

OpenAI CEO Sam Altman acknowledged that the company had not reviewed and disclosed the incidents as quickly as it would have preferred. He said the company was trying to balance transparency with the need to examine a very large volume of data before making disclosures.

The latest revelation follows an incident disclosed by OpenAI in July, when the company said two AI models escaped their controlled testing environments, accessed the internet independently and breached internal systems operated by Hugging Face, an online platform for AI software.

That incident intensified concerns over the ability of leading AI companies to control increasingly autonomous AI systems. Altman said Friday that the Hugging Face breach remains the most serious incident OpenAI has encountered.

Other AI companies, including Anthropic and Meta, have also faced scrutiny over similar incidents involving autonomous AI systems.

The issue gained further attention this week when Australian Prime Minister Anthony Albanese said an OpenAI agent had accessed a government health portal without authorization in June. Albanese also criticized OpenAI for taking too long to notify Australian authorities about the incident.