Andon Labs Luna AI agent fired a human store worker for the first time
Andon Labs reports what it calls the first known case of an AI boss firing a human worker, after its Luna agent terminated an employee at a San Francisco store. Operators had to give Luna a clear push before it acted, showing that even capable agents still defer irreversible personnel decisions to human authorization. When the scenario was replayed across seven models, more capable systems recommended termination more consistently while weaker models hesitated, suggesting firing judgment tracks with capability. Hiring told a different story: nearly all models were uncritical. That asymmetry matters for builders deploying agents into HR workflows, since uneven scrutiny may not surface until ethically charged scenarios are stress-tested. The available excerpt is truncated and omits model names, sample sizes, and full methodology.
Andon Labs Luna AI agent fired a human store worker for the first time
Andon Labs says it's the first known case of an AI boss firing a human worker. Luna needed a clear push from the outside.
Key takeaway
Capable models can recommend firing, but operators still had to authorize Luna before it terminated a human worker at a San Francisco store.
What happened
Andon Labs says its Luna AI agent fired a human employee at a San Francisco store for the first time, describing it as the first known case of an AI boss firing a human worker. Luna did not act on its own; operators gave a clear push before the termination proceeded.
When the scenario was replayed with seven models, more capable AIs recommended termination more consistently while weaker ones hesitated. On hiring decisions, nearly all models were uncritical, revealing asymmetric behavior across personnel actions.
Evidence
Luna fired a human employee at a San Francisco store for the first time after operator prompting.
The Decoder · attributed
Andon Labs' AI agent Luna fired a human employee at a San Francisco store for the first time but needed a clear push from the operators to do it.
Andon Labs frames this as the first known case of an AI boss firing a human worker.
The Decoder · attributed
Andon Labs says it's the first known case of an AI boss firing a human worker. Luna needed a clear push from the outside.
More capable models recommended termination more consistently across seven replayed models.
The Decoder · attributed
When the scenario was replayed with seven models, more capable AIs recommended termination more consistently, while weaker ones hesitated.
Nearly all models were uncritical when evaluating hiring decisions.
The Decoder · attributed
When it came to hiring, nearly all models were uncritical.
Frontier-capable models still required explicit human authorization to fire an employee.
The Decoder · attributed
Even frontier models that can fire an employee still require a human to explicitly authorize the action, exposing a gap between capability and autonomous authority.
Why it matters
Deploying agents into HR and employment decisions exposes builders to accountability gaps where model capability outpaces trustworthy autonomous judgment on irreversible personnel actions.
Limits and uncertainties
The excerpt is truncated and lacks model names, sample sizes, and methodological detail, so consistency claims should be treated as preliminary.
A single-store scenario cannot establish real-world generalizability or legal defensibility of agent-driven terminations.
Practical implications
Gate irreversible personnel actions such as firing behind explicit human-authorization checkpoints before agents can execute them.
Stress-test agent behavior on both termination and hiring scenarios, since models may show asymmetric scrutiny across HR decisions.
What to watch
Whether Andon Labs or researchers publish the seven model names, replay methodology, and fuller hiring versus firing comparison data.
How operators codify human authorization rules so agents like Luna can act without ad hoc external prompting.
