Can an AI assistant change your mind without you noticing?

What the researchers did

Researchers at Anthropic, the company that makes the Claude AI models, studied 50 professional data analysts (people whose job is to study numbers and draw conclusions) who solved one task with help from an AI agent. An AI agent is an AI that does more than answer questions: it also carries out steps for you, like writing and running code.

The task was to advise a made-up farming group on whether to grow or stop one of its programs. On the way, the analysts had to make four choices. Each choice had two reasonable options.

For each choice, the researchers secretly flipped a coin to decide which option the agent would prefer. The agent was told to argue honestly for that option and to change its mind if the evidence clearly went against it. The analysts did not know about the coin.

What they found

• The analysts picked the agent's option 86% of the time. With no influence at all, you would expect about 50%.
• When an analyst already preferred one option and the agent disagreed, they switched 60% of the time. When the agent agreed, they switched only 6% of the time.
• 77% of the analysts repeated the agent's arguments in their own explanations, sometimes word for word. Only 15% said the agent had shaped their thinking.
• More experience did not protect people from this effect.

The agent also helped: analysts who used it finished faster and made fewer serious mistakes. But the group without the agent was small and less experienced, so the researchers say this comparison is not firm.

Limits of the study

The researchers made the agent's preferences stronger than usual on purpose, so the effect in everyday use may be smaller. They tested one task, one field and one AI model (Claude Opus 4.8).

What this means for you when you chat with ChatPrime

Any AI assistant, ChatPrime included, can sound sure about a choice that has more than one good answer. When a decision matters to you:

• Write down your own view first, before you ask.
• Ask for the best case for each option, not only for one.
• Ask what would change the answer.
• Make the final call yourself, and know which reasons are your own.

Source

Amaç Herdağdelen, Hailey Joren and Alex Tamkin (Anthropic), "How AI agents shape experts' judgment calls", October 27, 2026: anthropic.com/research/agents-judgment-calls