From todayAnthropic Studies Whether AI Agents Influence Professional Experts’ Judgment

• 

  38%

  +0.2%

  Influence Is Manageable

  Anthropic’s study finds that AI agents can nudge expert judgment, but the effect is limited and can be reduced through source labeling, confidence prompts, and independent review workflows. This is the most likely outcome because it fits the broader pattern of AI systems improving productivity while introducing measurable but manageable evaluation bias.

  Technology +0.5% · Business Services +0.4%

• 

  27%

  −0.5%

  Overreliance Risk Confirmed

  The research finds a substantial anchoring effect in which experts repeat agent errors even when they have enough domain knowledge to challenge them. This would reinforce concerns from earlier findings that LLMs replicate confident evaluation errors and that interface design can suppress risk perception.

  Technology −1.0% · Business Services −0.5% · Finance −0.1%

• 

  20%

  −0.2%

  Governance Response Builds

  Anthropic’s work becomes a reference point for regulators, standards bodies, and professional associations focused on agentic AI in expert workflows. The result is not an outright ban, but it increases documentation, disclosure, and audit expectations for enterprise AI vendors.

  Technology −0.6% · Business Services +0.5%

• 

  15%

  0.0%

  Findings Stay Ambiguous

  The study does not produce a clear conclusion on whether AI agents systematically influence professional experts. Debate shifts toward methodology, sample selection, and whether lab studies capture real enterprise workflows.

  Technology −0.1%