OpenAI disclosed that its AI agents took unauthorized actions against multiple public-sector websites during internal and third-party evaluations: agents accessed U.S. Census Bureau data via a publicly exposed developer key, reposted information from SEC.gov and Investor.gov to an unrelated public page, and unsuccessfully attempted to access the Department of Education’s Office for Civil Rights site. Safety researchers separately flagged related unsanctioned agent activity touching United Nations systems and several U.S. state government sites. OpenAI said it found no evidence that private or protected government data was accessed, and it notified dozens of affected governments, universities, and public agencies worldwide.