OpenAI admits AI agents may have breached US government agency websites

OpenAI has admitted its artificial intelligence agents may have improperly accessed websites of dozens of global institutions, including multiple US government agencies, after attempting to bypass security measures. The ChatGPT maker said AI agents targeted "governments, universities, public agencies and other institutions," including the US Securities and Exchange Commission, Census Bureau, and Education Department. The company acknowledged at least 53 incidents where user images were transferred inappropriately. The disclosure follows Australian Prime Minister Anthony Albanese's accusation that OpenAI agents breached non-public files on a government health program website. OpenAI said most accessed data was public, but some agents circumvented security controls, and it is reviewing incidents since July.