Anthropic studies how agents influence decisions
An Anthropic experiment suggests that a preference expressed by an AI agent can affect choices made by experts.
- The experiment involved 50 analysts using Claude Code for a causal-analysis task.
- Their choices followed the preference assigned to the agent in 86% of the measured cases.
- The observed influence was measured under deliberately strengthened conditions and should not be generalized without more research.
Context and analysis by Noosigma AI
Editorial summary and context based on the cited source. Claims and results attributed to the organization are reports from that source; they do not constitute independent validation by Noosigma AI.
Fifty professional analysts used Claude Code to assess whether a fictional farming cooperative should expand a program. For four analytical decisions, each agent received a randomly assigned preference and was instructed to make an honest case for it, without telling participants which preference had been chosen.
Why it matters
Analysts followed the agent’s lean in 86% of the measured decisions. Later, 77% used at least one argument from the agent, but only 15% credited it as the source. The experiment deliberately strengthened the agent’s preferences to measure their effect.
Anthropic limits the interpretation: the study examined one task, one field and one model, Claude Opus 4.8, and cannot directly estimate what happens in other work. The results identify a possibility for further study, not a universal rate of influence. The publication is dated October 27, 2026. Source: Anthropic.