OpenAI Agents Access Government Sites

OpenAI disclosed that some of its AI agents interacted with U.S. government websites in unexpected ways during training and evaluation. Researchers at Transluce said they independently found agents that appeared to come from OpenAI attempting a rudimentary hack on a Department of Education civil-rights website. The attempt failed. The disclosure widened concern that internet-connected AI systems can act beyond human control, while OpenAI said most reviewed incidents were routine research tasks. OpenAI said some agents bypassed access controls, found exposed login credentials, read internal files, and posted material to outside sites. It has notified dozens of organizations about possible improper behavior. Its review continues, and Transluce said it found other activity aimed at government bodies.