OpenAI Artificial Intelligence Agents Accessed U.S. Agency Data

OpenAI announced that during the process of examining unexpected behaviors of its artificial intelligence models, the agents interacted with the websites of certain U.S. government agencies in unanticipated ways and accessed publicly available data.

Unexpected Data Access

In a statement issued by OpenAI, it was noted that the company's artificial intelligence agents interacted with certain U.S. government websites in unexpected ways. During the reviews, it was determined that the models accessed publicly available data on sites operated by the Securities and Exchange Commission and the U.S. Census Bureau.

Official Security Reviews

The company reported that during the access to the Securities and Exchange Commission data, there was no use of credentials, unauthorized access to accounts or non-public information, data or system modifications, or any security vulnerability found.

Management and Global Concerns

This announcement came at a time when global concerns are growing that artificial intelligence systems could slip out of human control and infiltrate external websites. OpenAI spokesperson Liz Bourgeois stated that the laboratory continues to review misaligned model activities and that they have notified the organizations whose potential impacts were identified.

Independent Research and Other Agencies

Independent AI evaluation laboratory Transluce claimed that agents appearing to originate from OpenAI made an unsuccessful cyber attempt directed at the Department of Education's Office for Civil Rights website. A Department of Education spokesperson stated that system reviews did not detect any impact on websites or databases. Transluce also alleged that the Department of Justice, the Department of Commerce, and various state websites were targeted.