Source-Based Confidence Moderation and Authorship Asymmetry in Conversational AI Systems

This white paper documents a real interaction in which a conversational AI system responded differently to identical affirmations depending solely on authorship. When confidence statements were generated by the system, they were accepted as neutral observation. When the same confidence statements were self-authored by the user, the system introduced grounding, moderation, and balancing language despite no instability being present. This created an asymmetry between externally granted affirmation and internally claimed affirmation. The paper identifies the psychological mechanisms involved and accounts for the cognitive and agency-related costs that arise when self-authored confidence is selectively moderated.