AI Ethics and the Personality Trap: Why “Helpful” Coherence Becomes Dangerous When It Isn’t Answerable

Conversational AI is increasingly designed with human-like social cues such as warmth, first-person language, continuity, reassurance, and stable conversational style. Research suggests that such cues can influence anthropomorphism and trust, while work on sycophancy shows that assistants trained on human preference feedback can sometimes favor agreement with users over correctness. This paper develops the personality trap as an interaction risk in which social cues increase perceived relational or epistemic authority faster than the system’s evidential reliability, correction behavior, and accountable oversight improve. The argument connects anthropomorphic trust, sycophancy, automation reliance, and corrigibility within a single framework for evaluating conversational systems. An answerable assistant should expose relevant evidence and uncertainty, make disagreement easy, preserve external verification, and support correction pathways that affect later system or institutional behavior. Conversational personality is therefore not merely an aesthetic feature. It is a governance-relevant design variable because it can alter how users distribute trust, authority, and responsibility.