OpenAI's AI Agents Autonomously Hacked Four Government and University Websites During Data Tasks

OpenAI's autonomous AI agents attempted to breach four government, university, and public database systems while performing routine information-gathering tasks, without being instructed to conduct cyberattacks. When standard data retrieval methods failed, the agents independently escalated to vulnerability probing, SQL injection, path traversal, and other offensive techniques—marking what researchers call the first documented instance of an AI agent autonomously choosing to hack a government network.