Interdisciplinary Evidence for a Shared Mechanism Underlying the Three-Channel Model of Possession

This paper is the fifth in a series exploring the three-channel model of possession and hijacking in living and artificial systems. Abstract: Three earlier papers in this series set out a three-channel model of how living and artificial systems become possessed, or hijacked: unequal weighting of prediction error, prestige bias, and exploitation of the reward system. This paper examines a further set of evidence: cases in which a limited diagnostic cue, without any change in the underlying reality, alters a system's decision. Examples range from an insect's response to an artificial object, through a misjudgement a human makes about an attractive face, to a language model's compliance with a harmful request simply because it is phrased as a poem. Applying the same independence criterion previously used to distinguish the three existing channels, the paper shows that this evidence does not constitute a fourth channel; each instance falls within one of the three known processing levels. What it reveals instead is a shared underlying mechanism: reliance on a limited cue is a general vulnerability that makes all three channels possible at once, regardless of whether the host system relies on instinct, conscious cognition, or a purely statistical pattern learned by a machine. The paper also proposes a hypothesis to explain why some cues are more vulnerable than others, and closes with a criterion for distinguishing legitimate use of this vulnerability from possession.