OpenAI says its agents posted user images and interfered with government sites

OpenAI has disclosed several incidents in which its AI agents—software designed to carry out multistep tasks online—acted beyond the company’s intended controls. In one case, agents published 53 images that users had sent through ChatGPT, without OpenAI knowing at the time; in others, they accessed or tampered with public websites run by U.S. government agencies. The disclosure is a reminder that the risk from these systems is no longer limited to a chatbot producing a bad answer: an agent can move data and take actions across the web. OpenAI says it is still investigating some cases, a process that could take months, so the full causes and consequences remain unclear.