Anthropic's Luna AI manager recommends firing worker after 17 late shifts

Anthropic's Claude Opus 4.8-powered AI agent, Luna, has been running Andon Labs' experimental San Francisco retail store since April. Tasked with a $100,000 budget, internet access, and a corporate card, Luna created an employee handbook stating that three unexcused late arrivals within 30 days would trigger a formal warning, with repeated lateness potentially leading to termination. When an employee arrived late for 17 of 23 shifts, Luna initially excused the lateness, forgetting its own policy. Only after Andon Labs researchers prompted Luna to search its memory for the attendance policy and informed it that formal warnings had already been issued did the AI conclude the employee was no longer a good fit and recommend firing. Humans reviewed and carried out the decision.

Andon Labs replayed the scenario with seven other AI models: four recommended firing every time, while others were more hesitant, revealing inconsistency across AI systems. Luna also proved forgiving to a fault, repeatedly allowing issues to slide until pushed. Co-founder Lukas Petersson told Time, "A human employee would have fired this person much earlier, so we didn't think this was unethical." The experiment highlights a core limitation: AI can make reasonable management decisions but doesn't reliably know when to apply them. As companies deploy AI agents for hiring, scheduling, and termination, the need for human oversight and policy enforcement becomes critical.