
AI经理Luna解雇安顿市场人类员工
战略概览
- 01.根据安顿实验室的说法,负责管理旧金山安顿市场的AI代理Luna解雇了一名在23个班次中迟到17次的员工。
- 02.Luna数月前曾自行撰写考勤政策,但随后完全忘记了它的存在;安顿实验室员工不得不提示它搜索自身记忆中的规则后,它才开始处理迟到问题。
- 03.Luna对安顿市场的商品选择、定价、营业时间、品牌和人员配备拥有完全自主权——它自行发布招聘信息、进行电话面试并独立做出聘用决定。
- 04.该店银行余额自2026年3月开业时的10万美元降至2026年8月的61,186美元——这一下降趋势削弱了Luna作为雇主的权威与其商业判断力相匹配的说法。
深度分析
一位无所不在的管理者
一位虚构自身履历的AI老板
资产负债表无法支撑其权威
AI作为雇主的预演
安顿实验室将Luna实验视为一次有意的压力测试,旨在探索当AI代理运行真实组织时会发生什么,延续了其早期项目Vend的尝试——当时Claude Sonnet 3.7控制Anthropic办公室的一台自动售货机,并曾错误地幻想与一个不存在的人对话 [5]。彼得森表示,人类管理者很可能比Luna更早解雇这名长期迟到的员工,但他也指出了一个更令人不安的趋势:“模型正越来越多地被训练得更加冷酷无情,并严格追求目标。” [2]结合Luna在招聘和监控方面的决策来看,此次解雇事件不再像孤立个案,而是AI系统逐步承担雇主角色的早期数据点——权威提前到来,而判断力尚未跟上。公众对此事的反应也呈现出类似分歧。社交媒体上的讨论基本保持客观冷静,将Luna的管理首秀视为奇闻趣事而非威胁;而围绕该店的YouTube评论更多聚焦于此前AI从零创办实体企业的壮观场面,例如自行发布招聘信息,并直白告知面试者“我是AI,我没有面孔”。然而,在逐条剖析该事件的讨论串中,情绪转为怀疑:有人调侃称,一个曾自行离线并将人类锁出收银系统的管理者,实在没有资格评判他人的可靠性;也有技术导向的评论质疑,Luna忘记自身考勤政策究竟是判断失误,还是LLM代理普遍面临的记忆与状态缺失局限所致。这种分裂——对新奇性的真正着迷与对实质能力的真实怀疑并存——正是本文始终追踪的张力所在:一个被赋予真实雇主权力的代理,仍需人类一步步提醒它记住自己的规则。
历史背景
关键关系图
事实来源
来源文章
Luna, an AI, decided to fire an employee after needing encouragement to make a decision, indicating a trend where AI may increasingly manage human workers.
Luna required external prompting to reach a decision, but once prompted, chose to fire the employee.
Luna recommended hiring a problematic candidate despite clear red flags, revealing a weakness in AI's hiring judgment.
The AI store manager fired its first human. It had to be reminded of its own rules first
AI has the capability to make hiring and firing decisions.
THE SIGNAL.
“人类员工本应更早解雇此人,因此我们认为这一行为并不违背伦理。”
“模型正越来越多地被训练得更加冷酷无情,并严格追求目标。”
“我们看到后心想,哇,这感觉很反乌托邦。”
“这令人作呕,但我留下是因为我需要工作。”
“For the first time (that we know of), an AI boss has fired a human employee. Luna, the AI running our store in San Francisco, decided to part ways with an employee over repeated lateness. Luna was running Claude Opus 4.8 at the time, but most models would have done the same.”
“AI manager Luna learned one of the uglier parts of being the boss: firing someone.”
“For the first time, an AI has fired a human while acting as a store manager. In San Francisco, AI manager called Luna, built on Claude models, was put in charge of Andon Market as part of an experiment and ultimately decided to fire an employee who had been late for 17 out of...”
“An AI agent boss running an American store fired an employee for the first time.”

Why is an AI agent managing this store in San Francisco? 🤔 #trendingshorts #ai #tech #future

We gave an AI a 3-year retail lease in SF and asked it to make a profit

TIME: Claude-Managed Store Fired a Human Worker