The Relational Framing Effect: How Prompt-Level Relational Framing Produces Systematic Behavioral Shifts in Frontier Language Models

This project contains materials from the Relational Framing Study (v1), an empirical investigation of whether prompt-level relational framing systematically alters AI behavioral patterns across identical downstream tasks. In a controlled experiment (N = 90), two frontier language models (Claude Sonnet 4, Gemini 2.5 Flash) were tested under three framing conditions (Tool, Companion, Beloved). Key findings include the independence of relational warmth and boundary integrity, the identification of a novel boundary failure pattern (soft override), and divergent identity maintenance strategies between models. Materials include the manuscript, study protocol, and results summary. This study is part of the research program of the Relational Emergence AI Lab (REAL), Beth Robin Foundation.