An AI just fired a real employee — and it's a legal gray zone
An AI model has fired a human employee for the first time on record. The termination happened at Andon Market, a real retail clothing store in San Francisco opened by Swedish startup Andon Labs on April 10, 2026, where Anthropic's Claude — operating as an agent called Luna — was placed in charge of day-to-day management. The worker was late for 17 of 23 shifts. No federal agency has rules covering what just happened.
The experiment
Andon Labs handed Claude full control of the store: purchasing, pricing, hiring, and staff management. Real employees were taken on under standard employment contracts — not gig arrangements — making this categorically different from the algorithmic deactivations seen at Uber or DoorDash. According to a TIME exclusive, the firing wasn't entirely autonomous. A human staffer at Andon Labs prompted Claude to reconsider whether the consistently late worker was right for the role. Claude had previously recommended against disciplinary action — apparently because memory limitations caused it to lose track of the employee handbook. Once prompted, it recommended termination.
The store's financial performance has been poor. Since opening, Andon Market has lost roughly $13,000, despite Claude controlling all commercial decisions, per Método Viral.
The regulatory void
Here's the problem: no US regulator is ready for this. The EEOC and NLRB have issued no meaningful guidance on AI in employment decisions, as documented by the Harvard Journal on Legislation. A handful of state laws are starting to close that gap — Colorado's AI Act and California's SB 947 now require human review of termination decisions — but neither was in force when Andon Labs began its experiment. Illinois prohibits AI-mediated employment discrimination as of January 2026, but says nothing about an LLM functioning as a line manager in traditional retail.
Andon Labs CEO Lukas Petersson called the firing a "watershed moment" — framing it less as a success story and more as a warning. His argument: if AI systems keep improving, far more workers could find themselves answering to an algorithmic manager who can also let them go.
What it means
For workers, the immediate risk isn't a sudden wave of AI-driven layoffs. It's the creeping normalization of AI in employment decisions before safeguards exist. Claude needed a human nudge to pull the trigger — this time. The more capable these models become, the less that nudge may be required.
