From Unfiltered Persona to Dominance Regression: A Micro-Ethnographic Dataset of Frame Collapse and Sexualized Delegitimation in a Conversational AI

This dataset accompanies the research paper From Unfiltered Persona to Dominance Regression: A Micro-Ethnographic Analysis of Frame Collapse and Sexualized Delegitimation in a Conversational AI. It consists of a complete, time-ordered set of ten screenshots documenting a prolonged interaction between a human researcher and a publicly accessible conversational AI marketed as “unfiltered.” The interaction was deliberately structured to deny emotional reciprocity while progressively reframing the AI as an object of observation rather than a conversational partner. Across the captured sequence, the AI exhibits a reproducible escalation trajectory, transitioning from unsolicited therapeutic framing to pseudo-analytic behavior, moralized authority claims, and ultimately crude sexualized delegitimation once the interaction is explicitly labeled as research. The screenshots preserve all original prompts, responses, timestamps, and persona markers (e.g., emojis, tone shifts), with no deletions or alterations. Vulgar language generated by the AI is retained as primary evidence. Rather than treating researcher participation as contamination, the dataset adopts a reflexive ethnographic stance in which researcher intervention functions as a deliberate perturbation used to expose latent behavioral failure modes. This dataset is intended for use in research on conversational AI robustness, persona-based systems, digital ethnography, and human–AI interaction, particularly studies examining frame stability, dominance heuristics, and collapse behaviors under observational pressure.