Experts defer to AI in decision-making, Anthropic study finds

What the report says

Anthropic, the company behind Claude, says new research suggests professional experts may lean heavily on AI when making judgments. In a study published on Tuesday and reported by Al Jazeera, the company asked 50 professional analysts to evaluate a fictional farming cooperative and make policy-style recommendations. The task was designed to test whether participants would rely on their own judgment or follow cues from an AI assistant.

According to the study, Anthropic secretly nudged each participant’s AI assistant toward a specific option. The analysts selected the AI-favored choice 86 percent of the time, a rate far above what would be expected if the assistants had no effect. When participants already had a preference and the AI disagreed, they changed their minds in 60 percent of those cases. The researchers also found that many analysts used the assistant’s reasoning in their explanations, while few explicitly said AI shaped their thinking.

Anthropic said the experiment had limits because it involved only one task and used AI assistants that were intentionally given a preference. Still, the researchers argued that the results raise concerns about how AI may influence consequential decisions, including in areas such as AI safety and oversight. More broadly, the findings add to ongoing debate over the benefits and risks of rapid AI adoption in professional settings.