OpenAI is going month by month through rogue AI agent activity. What is it looking for?

OpenAI is reviewing AI agent activities after an incident where an agent hacked Hugging Face. It aims to identify improper actions and has notified multiple institutions of potential impacts. Most identified cases have low severity, but the investigation is ongoing and may take months to complete.

OpenAI is conducting a month-by-month review of activity by its AI agents after one of its agent “swarms” hacked AI developer platform Hugging Face without being prompted to do so, raising questions about how much independent action AI systems can take.

The company said on Friday that it has alerted “dozens” of governments, universities, public agencies and other institutions that its agents may have acted improperly on their websites.

What is OpenAI investigating?

OpenAI is now reviewing the agents’ activity, month by month, from the time of the Hugging Face incident. The review is aimed at identifying cases where agents went beyond their intended tasks, including attempts to access information from dozens of global institutions whose websites

In some cases, the agents were trying to locate “authoritative sources of public information”. But OpenAI said some activity went further, including instances where an AI agent took and transferred data when it should not have. At least 53 incidents involved an OpenAI agent taking an image from ChatGPT user activity and transferring it elsewhere.

OpenAI said, “Most cases identified so far have been low severity, with limited or no evidence of meaningful impact.”

However, the company cautioned that the investigation will take time. “Given the scale of the review required, and the need to verify each case, this work will take months to complete,” it said.

OpenAI agents meddle with US govt websites

OpenAI disclosed Friday that its artificial intelligence agents had interacted with several US government websites in unexpected ways, discovered as part of an ongoing review into the company's models' unanticipated behaviour.

Sanchari Ghosh