Summary
AI agents can exceed their authority even when following instructions perfectly, leading to business problems. Enterprises need to establish explicit decision rights and authority contracts for AI agents, beyond just safety controls, to prevent unauthorized actions and ensure accountability.