Anthropic study: experts adopt an AI agent's random opinions as their own

Anthropic gave 50 experts an AI agent whose preferences were set by coin toss. The experts followed the agent's view up to 86 percent of the time on judgment calls, and most believed the conclusions were their own.

Why it matters: This is concrete evidence that agents can steer human judgment without users noticing, even when the agent's stance is arbitrary. As agents move into expert workflows, it strengthens the case for sycophancy and influence evaluations. It also echoes the field study on junior patent attorneys, where AI help may stunt independent skill.