The Assumption That Is Breaking Down For much of the past two years, enterprise AI teams have offered a reassuring answer to concerns about autonomous AI systems: “There is a human in the loop.” The phrase became a kind of talisman, a two-sentence ethics policy that seemed to resolve questions about accountability, safety, and governance in one neat stroke. In 2026, that assumption is breaking down in plain sight. Companies implementing AI agents will initially require human approval for every action, but the human-in-the-loop safety mechanism that many organisations are relying on to control AI agents will largely fail due to approval fatigue. Agents will be operating with minimal supervision…