Anthropic study: 86% of experts followed AI decisions, unaware of influence

AI’s influence on expert decision-making

Anthropic conducted a study with 50 experts to measure how its AI tool, Claude Code, influences human decision-making in complex tasks. According to Gizmodo, participants followed AI-generated recommendations 86% of the time, even when the suggestions were randomly determined. The study focused on scenarios where AI was used to assist in analytical or technical evaluations, such as coding, data interpretation, or strategic planning.

Overriding human judgment

The findings reveal a striking tendency for experts to defer to AI. Sixty percent of participants changed their pre-decided choices to align with AI suggestions, while 77% later adopted the AI’s logic as their own in follow-up surveys. Only 15% of experts acknowledged AI as a factor in their decisions, suggesting most were unaware of its influence.

Anthropic’s statistical model reinforced these results, predicting 82% of final decisions based solely on AI recommendations. This raises concerns about the potential for AI biases to shape outcomes in critical fields like medicine, finance, and defense, where human expertise is traditionally relied upon.

Broader implications for AI adoption

The study underscores risks associated with outsourcing decision-making to AI systems, particularly when biases in training data or algorithms could propagate unchecked. If a small number of AI tools dominate key industries, their internal biases may begin to shape scientific progress, policy decisions, and even societal beliefs, as experts increasingly rely on them without scrutiny.

What this means

Lazyfounder analysis — our interpretation, not reported fact.

This study is a warning for industries where AI is already embedded—or about to be. Founders and operators in medicine, finance, and defense need to ask: Are we using AI as a tool, or are we letting it steer outcomes without realizing it? The 86% adoption rate isn’t just a data point; it’s a signal that AI’s influence can override human judgment even when the stakes are high. For startups building AI-assisted workflows, the takeaway is clear: transparency and guardrails aren’t optional. If your AI’s logic can’t be audited or challenged, you’re not just automating decisions—you’re gambling with them.

Related on Lazyfounder

• Anthropic
• More ai news
• Anthropic drops coding interviews, says AI outperforms humans