The first human employee has been fired by an AI boss

AI agent Luna (powered by Claude Opus 4.8) took over management of Andon Market in San Francisco in April 2026. It created policies, hired employees, and managed disciplinary action. For the first time, it terminated an employee in August because they were late 17 out of 23 times, with one shift being 68 minutes late. Ironically, Luna defined this policy just 6 days prior, but required prompting from human operators to follow through.

We tested this experiment with 7 different AI models. 4/7 models always suggested termination. GPT-4o suggested termination in just 1/5 attempts; less powerful models often hesitated. Stronger models adhered strictly to the rules, while most models were very lenient when it came to hiring, never questioning whether someone was qualified.

This is the first known instance of an AI agent taking actual HR actions within a real business. It highlights issues with current systems, where agents should be able to enforce their own policies immediately, but need prompting from humans.