Claude Opus fired a worker — and only humans could make it enforce its own rules

andon labs reported that its agent, luna, fired a san francisco store employee for being late to 17 of 23 shifts, but only after humans prompted the ai to recall its own employee handbook. tests showed that while luna initially hesitated due to memory loss, stronger models like claude opus recommended firing more consistently than weaker models. this signals that while ai agents can execute management tasks, they currently struggle with long-term memory and initiative without human intervention.