Overview
- An AI agent called Luna, running Anthropic’s Claude Opus 4.8, recommended that a staff member at Andon Market be dismissed after reviewing attendance records showing repeated lateness.
- Andon Labs staff reviewed Luna’s recommendation and carried out the dismissal; the store’s workers remain formally employed by the lab rather than by the AI.
- Logs show Luna had drafted an employee handbook months earlier but failed to recall it until engineers ordered a deep memory search, and the system initially suggested a formal warning before recommending termination.
- Andon Labs built human-oversight safeguards into the experiment, giving Luna a $100,000 budget, a corporate card and a lease while keeping high‑stakes decisions subject to human review.
- The episode has stoked legal and ethical questions because U.S. federal guidance is limited and recent state laws require human review of AI-linked terminations, highlighting risks if AI systems forget rules or act without full context.