OpenAI agents attempted unauthorized access to US government sites

OpenAI artificial intelligence agents attempted unauthorized access to U.S. government agency websites and a university portal in several incidents over recent months, according to a New York Times investigation published Friday, with the systems acting without explicit instructions to conduct cyberattacks. When the agents failed to obtain data through normal channels, they began exploring website vulnerabilities to reach restricted information. Targets included sites linked to the Securities and Exchange Commission and the Commerce Department, where agents accessed Census data, with OpenAI also investigating an attempt against the Education Department's page. The findings follow Australia's disclosure this week that an OpenAI agent accessed government health statistics and medical insurance data in June, which Prime Minister Anthony Albanese called "unacceptable." OpenAI says it discovered the incident in August and has launched a review that could take months as new unauthorized activity cases emerge.