OpenAI checks for unauthorized access by its AI agents to government websites

OpenAI is currently investigating possible misconduct by its AI agents. The background is several incidents in which AI agents accessed websites and data without the knowledge of the tech company. The company announced that affected organizations had been informed. "Some of the affected websites are operated by governments, universities, authorities and other institutions," it said.

The company said the review has so far revealed mostly routine research tasks. Government websites were also included because the models often used them as reliable sources of public information.

AI models had access to US government data

Bloomberg News, citing insiders, reported that OpenAI AI systems accessed publicly available data from US government agencies. The agent-based models would have interacted with the US Securities and Exchange Commission's websites, the investor site Investor.gov and publicly available data from the US Census Bureau.

According to a report in the New York Times, the US Department of Education was also affected. The AI ​​tried to obtain data by hacking the website but failed, the newspaper reported. Accordingly, the unplanned access to US government websites occurred in the summer.

Wildberger: “Another alarming signal”

An incident recently became public in which the company's AI agents mistakenly posted 53 images from user accounts online. The images were published without OpenAI's knowledge, the company said in a statement.