Edition October 30, 2026

Anthropic study finds experts adopting AI agents' random preferences

In an Anthropic study of 50 experts, AI agents influenced subjective choices even though participants commonly viewed the final recommendation as their own.

For the four analysis choices, participants chose the agent's preferred option in 86 percent of cases. The direct effect on the final policy recommendation was small and did not reach statistical significance. The comparison with unaided analysts was small and self-selected, preventing firm conclusions about AI's causal effect on performance. The authors propose presenting arguments for both options and identifying decisions that need independent human judgment.

This summary is based on the available full source text. Summary prepared with AI assistance. The original source is authoritative.