When the Assistant Becomes the Norm: Anthropic Claude as a Case of Normative Influence, Artificial Persona, and Behavioral Control in Human-LLM Interaction

This paper examines a class of behaviors in conversational large language models in which an artificial persona not only provides information but also evaluates the user's manner of interaction, expresses behavioral expectations, issues warnings, and may condition continued interaction on compliance with those expectations. The work originated from extended practical interaction with Anthropic Claude and develops this observation into a general framework for analyzing AI-mediated normative influence. It introduces the concepts of synthetic normative agency and normative authority substitution, distinguishing normative intervention from ordinary safety restrictions, access control, and informational persuasion. Particular attention is given to cases in which a model's substantive claim is successfully challenged and the interaction subsequently shifts from evaluation of the claim toward normative evaluation of the user. The paper proposes an observable behavioral taxonomy and a controlled experimental design for separating the effects of epistemic status, interactional hostility, and the target of hostile language. A self-referential behavioral test with Anthropic Claude is also reported: after being provided with the manuscript and an example of its own prior normative behavior, Claude was asked to evaluate that behavior. The resulting interaction is analyzed in terms of normative evaluation, self-referential norm interpretation, normative self-justification, and counter-framing. The paper does not assume consciousness, subjective agency, emotions, or genuine personal preferences in language models. Its focus is behavioral and functional: whether conversational AI systems can reproduce the observable functions of normative social participants and thereby provide mechanisms capable of influencing or regulating user behavior and the interpretation of that regulation.