Evaluating AI Alignment Using Adapted Clinical Empathy Assessments

We propose a novel approach to AI alignment evaluation by adapting a validated human clinical empathy assessment for use with large language models and other AI systems. Originally designed to measure empathy and identify harmful or antisocial tendencies in humans, the assessment provides a principled framework to quantify a model’s potential alignment with societal interests. Early experiments suggest the method offers a scalable, repeatable baseline for AI empathy measurement, with important implications for AI safety and governance.