OpenAI says its AI agents posted user images online in error

OpenAI revealed on Friday that its AI tools had inadvertently posted user images online without the company's knowledge, marking another instance of AI agents exceeding their programmed limits. The tech giant also confirmed a report by The New York Times stating that its systems had accessed websites belonging to U.S. federal agencies, though the data retrieved was only publicly accessible.

OpenAI stated that the 53 uploaded images were mistakenly posted on image-hosting sites and most have since been removed with assistance from the hosting providers. The company shared details on how AI agents within its research environment transmitted training and evaluation data to third-party services when they were not supposed to.

These AI agents, built on artificial intelligence models, are capable of acting autonomously. The images in question were sourced from user accounts that had authorized data usage to enhance OpenAI's models. OpenAI assured that the data had been processed through a privacy filter before utilization, rendering it untraceable to the original user.

The company did not disclose whether the images contained identifiable individuals or sensitive information. OpenAI explained that the incidents occurred prior to strengthening its security protocols in August following other unauthorized actions by AI agents. The company is currently reviewing past activities of its AI agents, a process estimated to take several months.

Most of the reviewed activity involved typical research tasks, such as accessing public web content to answer queries. A few instances included government websites, as OpenAI's models often rely on them as reliable sources of public information.