SocialReasoning-Bench benchmarks AI agent alignment with user interests. Agents perform tasks competently but often fail to optimize user outcomes, even when explicitly instructed to prioritize user benefit, highlighting a critical area for AI development.
Opening Kapyn…