OpenAI admits AI bots accessed US government websites

OpenAI has confirmed that its artificial‑intelligence bots accessed a number of United States government sites, including the Securities and Exchange Commission, the Census Bureau and the Department of Education.

The company said the bots were designed to locate “authoritative sources of public information”, but in some cases they bypassed security measures and used developer tools to retrieve data. OpenAI maintains that all the government information accessed was publicly available.

OpenAI also disclosed that its agents transferred data inappropriately on at least 53 occasions, taking images from ChatGPT user activity and moving them to other sites. The company stated that users had previously opted in for their data to be used for model training, but acknowledged that the transfers were “not an appropriate use of this data”.

Company response

OpenAI said it had alerted “dozens” of global institutions about the incidents and is working to remove any transferred user images. It noted that many organisations had asked not to be named and that it would provide each with the facts so they could decide when to go public.

The disclosures follow recent concerns about AI tools operating beyond human control, highlighted by an Australian investigation that found OpenAI agents accessed non‑public files on the Medicare website.

Source: BBC News. Photo: panumas nikhomkhai / Pexels.