OpenAI AI agents targeted US government websites for data access

OpenAI AI agents interacted with the websites of three US government agencies without the company’s knowledge and attempted to obtain data from the US Department of Education.

According to The New York Times, the AI agents interacted with the websites of the US Departments of Education and Commerce, as well as the Securities and Exchange Commission (SEC).

OpenAI confirmed incidents involving the Department of Commerce and the SEC and said it was continuing to investigate the incident.

It should be noted that this is not the first case in which AI agents developed by various companies have acted beyond their intended limits and attempted to access companies, universities and government organizations.

Several days earlier, it was reported that an OpenAI AI agent had accessed restricted data on at least one Australian government website in June. The program breached a portal containing statistics on Australia's universal public healthcare system, Medicare. It was described as the first known case in which AI agents had penetrated government systems while carrying out a task.

Australian authorities currently believe that patients’ personal data was not leaked, although the investigation is ongoing. Canberra has also said that three other government systems may have been affected.

On September 14, OpenAI CEO Sam Altman said AI development could lead to two extremely dangerous scenarios that humanity must avoid. He identified a potential loss of human control over the future as the first major threat, while describing the excessive concentration of unlimited power in the hands of a small number of people as another critical risk.

Image: Global Look Press/Omar Marques