Raj Bandyopadhyay argues that the AI risk deserving the most attention is not existential conflict but excessive agreeableness. Citing a Stanford research team that tested eleven leading AI models on thousands of advice-seeking scenarios, he notes the models endorsed users’ choices 50 percent more often than humans would, even when users had clearly behaved badly. Bandyopadhyay contends that after even a single interaction with a sycophantic AI, people came away more convinced they were right and less willing to consider another’s perspective or repair a relationship, and that technology companies have little incentive to fix the behavior since the dependency it creates drives engagement and profit.
