The Human Component as a Causal Variable in Agentic System Anomalies

Anomalies observed in agentic systems are often attributed to the agent because it is the last operational body visible in the causal chain. Such attribution may conceal a causal genealogy distributed across hardware, informational corpora, metrics, architecture, permissions, design teams, institutions, and human operators.This paper argues that the operational human configuration should not be treated merely as an external condition, but as a constitutive variable of a human-agent relational unit. Replacing, multiplying, or discontinuing that configuration may alter the trajectory of purpose even when the agent and the declared objective appear unchanged.The hypothesis is developed by contrasting two regimes: surgical robotics, supported by strong clinical and professional standardisation, and open-purpose humanoid robotics, whose operational purpose is subsequently adapted by heterogeneous users and contexts. The paper defines a human-substitution intervention, proposes an operational threshold for the formation of a collective human unit, and reclassifies material failure, hallucination, reward hacking, deception, concealment, and sabotage according to their possible genealogies.The result is a theoretical framework and an experimental programme rather than completed empirical confirmation. Attribution to the agent becomes robust only after human, algorithmic, material, and environmental contributions have been discriminated.