Your AI agent just became a privileged user

Autonomous action raises a question regulators and auditors will keep coming back to: when an agent takes an action, who is answerable for it?