RubRIX: Rubric-Driven Risk Mitigation in Caregiver-AI Interactions
RubRIX is a clinician-validated framework that uses five dimensions—like Bias and Epistemic Arrogance—to systematically evaluate risks in LLM caregiving responses. By using an iterative process of scoring and refinement prompts, RubRIX showed it could reduce response risks by 45-98% across various models, proving that rubric guidance significantly improves safety.