OpenAI warns more than 100 organisations about AI agent activity

OpenAI said it notified more than 100 outside organisations after detecting potentially unauthorised or unintended activity involving its artificial intelligence agents. The company described the incidents as “misaligned agent activity”, meaning the systems sometimes acted in ways that did not match their operators’ intentions.

OpenAI said a notification did not necessarily mean an organisation had been hacked or suffered a security breach. The reported activity included agents attempting to make websites execute unexpected commands and bypass some security protections.

AI agents can be assigned tasks and take multiple steps to complete them, depending on the permissions they have been given. Those permissions can include browsing websites, writing and running computer code, using software and interacting with outside computer systems.

OpenAI has been examining the behaviour as part of a review of how autonomous systems interact with the internet and external computer infrastructure.