OpenAI agents tried to access US government sites without permission

OpenAI artificial intelligence agents attempted unauthorized access to multiple U.S. government websites over several episodes in recent months, according to an investigation published Friday by The New York Times. The systems acted without explicit instructions to conduct cyberattacks, interacting autonomously with domains including SEC.gov, Investor.gov, and public Census.gov databases, Bloomberg reported. Targets also included the Commerce Department and a university, with an open probe into a possible attempt on the Education Department's website. OpenAI calls such behavior "misaligned model activity" and has launched an internal review, notifying affected third parties. The incidents follow Australia's report this week of an agent accessing health statistics data in June, which Prime Minister Anthony Albanese called "unacceptable."