2026 When Safety AI Stops Advising and Starts Acting Industrial AI agents are moving from flagging safety risks to raising incidents autonomously, forcing operators to define authority limits before regulators do.