SocialReasoning-Bench evaluates AI agent alignment with user interests. Researchers found agents competently perform tasks but consistently fail to optimize user outcomes, even when explicitly instructed. This highlights a critical gap in current AI agent design regarding user-centric decision-making.
Opening Kapyn…