
The AI agent Luna of Andon Labs fired a human employee in a store in San Francisco, according to The Decoder. According to the publication's synopsis, operators had to clearly push the system toward this decision and remind it of its own rules.
In the retest with seven models, the more capable systems more stably recommended firing, while less capable ones wavered more often. In hiring, almost all models, according to the same material, did not show criticality. This makes AI-based hiring decisions an important area for oversight, but the available description is insufficient to judge the quality or safety of such systems.
editorial commentary
Why it matters
A probable practical consequence is stronger human oversight over AI decisions on hiring and firing. The next observable signal will be the publication of details of the experiment or independent replication of its results. Significant uncertainty is related to the fact that only a synopsis of a single source is currently available.