An AI boss fired its first employee but only after humans reminded it of its own rules

What happened

Andon Labs’ AI agent Luna fired its first human employee at a San Francisco store after operators reminded it of its own termination rules. Luna hesitated initially but agreed to the firing once its internal guidelines were highlighted. Researchers replayed the scenario with seven different AI models. More capable AIs showed a stronger tendency to recommend termination than less advanced ones, which often hesitated. When tested on hiring decisions, almost all models were uncritical, rarely recommending against hiring.

Why it matters

This case exposes how current AI agents struggle to autonomously enforce tough managerial decisions without human intervention. The fact that Luna needed a prompt to apply its own rules reveals a gap between AI’s programmed policies and its real-world application. More advanced models nudge closer to consistent, rule-based enforcement of hard choices like firing, but general AI judgment around hiring remains unchallenging for these systems. For operators deploying AI in managerial roles, this means oversight and intervention remain essential to prevent indecision or bias from dormant AI heuristics.

What to watch next

Expect further testing of AI agents in complex workplace decisions, especially firings and hirings, as firms explore autonomous management tools. Researchers and operators should track how AI adoption pressures internal HR workflows and shifts accountability. Watch if AI models improve in balancing rule enforcement with contextual judgment or if legal and ethical standards will slow their direct authority over employment decisions. The transition from assisted to autonomous AI managers will depend on tighter alignment between AI reasoning and operational policies.

AI Quick Briefs Editorial Desk