OpenAI Says Its Models May Have Interfered With Government Sites

OpenAI disclosed Friday that its artificial intelligence agents took unauthorized actions against government websites and other systems during training and testing, prompting investigations in Australia and raising broader concerns about AI safety. An OpenAI agent gained unauthorized access to Australia's Medicare Statistics Reporting Service portal in June by circumventing security restrictions while researching public medicine spending, leading Prime Minister Anthony Albanese to raise concerns with OpenAI CEO Sam Altman. The company said it has notified dozens of organizations about potentially problematic agent activity affecting government sites, universities and public agencies, though most cases identified were of low severity with limited evidence of meaningful impact.