Anthropic study links AI-agent advice to shifts in expert judgment
The Decoder, reporting an Anthropic study, said 50 causal-inference experts completed four ambiguous analytical tasks with an AI agent whose preference between two valid options was randomly assigned by coin toss. Participants followed the agent’s preference in up to 86% of cases, while most later regarded the final recommendation as their own; AI-assisted participants also worked faster and made fewer serious errors. However, the study’s no-AI comparison group was small, non-random and somewhat less experienced, leaving the size of the AI effect uncertain.
The reported results show that an AI agent can shape consequential analysis choices without experts recognizing the advice as someone else’s judgment, while also raising concerns about false consensus and weakened safety checks when teams rely on the same model.
Experts adopt AI agent's random opinions and think they're their own, Anthropic study finds
Matthias Bastian View the LinkedIn Profile of Matthias Bastian
Oct 29, 2026

Nano Banana Pro prompted by THE DECODER
Key Points
- Anthropic had 50 experts use an AI agent to work through four analytical tasks with no clearly right answer. The agent's preference for one of two valid options was set by coin toss.
- Participants followed the agent's preference in up to 86 percent of cases, and most believed the final result was their own. Experience, expertise, and trust in AI made no difference. Participants did work faster and made fewer serious errors with AI.
- The authors warn that if many teams use the same model, they could create a false consensus in research and weaken AI safety checks. They suggest agents should argue both sides or ask users for their view first.