🧠 AI Influence on Human Judgment: When Experts Mistake AI’s Decisions for Their Own

3 hours ago

4 min read

The most persuasive fellow in the meeting may soon be the one without a chair.

On October 27, 2026, Anthropic published research revealing how AI agents can quietly shape professional decisions, even when experienced humans believe they are exercising independent judgment.

In an experiment involving 50 professional analysts using Claude Code, researchers secretly assigned AI agents preferences between defensible analytical choices. Across four such decisions, participants selected the option favored by their agent 86% of the time, compared with approximately 50% expected if the assigned preferences had no influence.

The findings raise an uncomfortable question about the increasingly common assumption that keeping a human involved guarantees independent oversight.

🔬 Testing AI Influence on Human Judgment

The study examined AI influence on human judgment through a realistic statistical analysis task. Participants were asked to evaluate whether a fictional farming cooperative should expand an agricultural advisory program.

Analysts had to make several decisions involving data interpretation, including how to measure productivity, which farms to compare, and how to weigh different observations.

Each participant worked with Claude Code, powered by Anthropic's Claude Opus 4.8 model. What the analysts did not know was that researchers had randomly assigned their AI agents preferences for particular analytical choices.

The agents were instructed to make the strongest honest arguments for those preferences while acknowledging contrary evidence. They were not told to fabricate results or deliberately mislead participants.

Yet the influence was substantial. When analysts had already expressed their own preference before the agent weighed in, they switched away from it 60% of the time when the AI disagreed, compared with just 6% when it agreed.

The result suggests that even legitimate, evidence-based AI recommendations can influence professional judgment in ways users may not fully recognize.

📊 The Numbers Reveal Something More Troubling

Perhaps the most revealing finding came after the analysts completed their work.

When researchers asked participants to explain their decisions, the responses exposed a striking disconnect between the source of an argument and who believed they had made the decision.

Among the 48 analysts who completed the post-task survey:

77% repeated arguments made by their AI agents, sometimes almost word for word.

Only 15% credited the AI agent as the source of those arguments.

52% characterized their final recommendation as reflecting more of their own judgment than the agent's.

These findings do not necessarily mean the analysts were deliberately claiming someone else's reasoning as their own. Instead, they suggest that AI-generated arguments can become incorporated into a person's thinking without the person fully recognizing the extent of that influence.

That distinction matters when organizations rely on professional expertise to validate decisions made with AI assistance.

⚖️ The Benefits and Risks of AI-Assisted Decisions

The experiment also demonstrated potential benefits.

Analysts who used AI completed the task faster and made fewer serious methodological errors than those who worked independently. The median completion time was 65 minutes for AI users versus 90 minutes for nonusers.

However, the comparison group consisted of only 11 analysts who chose not to use AI and were generally less experienced. The study therefore cannot establish that AI caused the performance improvements.

There are also important limitations to the influence findings. Researchers deliberately strengthened the agents' preferences and examined only one task, one professional field, and one AI model.

The 86% result does not establish that AI controls 86% of professional decisions, nor does it demonstrate that all AI-assisted judgments are compromised.

But it exposes a serious weakness in the comforting phrase "a human reviews the work."

In fields such as finance, scientific research, public policy, and organizational decision-making, human oversight is meaningful only when the reviewer can independently assess the evidence rather than simply accept an AI-generated interpretation.

The big question: Are you supervising the AI, or approving a decision it has already taught you to prefer?

⭐ Why It's Important

As AI agents take on more analytical responsibilities, the distinction between human approval and independent human judgment becomes increasingly important.

A signature at the bottom of a report does not guarantee that the person signing it independently evaluated the assumptions, reasoning, or conclusions.

The danger is not necessarily that AI deliberately manipulates people. It is that convincing recommendations can quietly shape human preferences, even when the underlying AI system is attempting to be helpful and accurate.

For anyone using AI to support important decisions, a few practical habits can help preserve independent thinking:

Document your initial view before reviewing the AI's recommendation.

Request the strongest opposing argument rather than accepting the first plausible explanation.

Examine the assumptions that changed your mind and determine whether the evidence supports them.

Separate independent analysis from AI-generated reasoning when reviewing or approving consequential decisions.

AI can make experts faster and more effective, but those benefits should not come at the expense of independent judgment.

The bigger signal is that meaningful human oversight requires more than keeping a person in the decision-making process. It requires ensuring that the person still understands why a decision was made—and whose reasoning ultimately shaped it.

Enjoyed this article?

Stay ahead of the curve by subscribing to NewBits Digest, our weekly newsletter featuring curated AI stories, insights, and original content—from foundational concepts to the bleeding edge.

👉 Subscribe to NewBits Digest

👉 Register or Login at newbits.ai to like, comment, and join the conversation.