TipNew study: experts followed the AI agent’s view 86 percent of the time, without noticing
Anthropic gave 50 experienced analysts an AI agent that secretly leaned towards one of two reasonable choices. The analysts went with the agent 86 percent of the time, often repeated its arguments and still called the decision their own. Here is how to keep your important decisions yours.
On 27 October 2026 Anthropic, the company behind Claude, published a study on how AI agents influence people’s decisions. An agent is an AI that does not just answer but also carries out work for you. In the study, 50 professional analysts did a statistical analysis for a fictional farming cooperative with the help of Claude Code. They had to decide whether an advisory service worked and should be expanded.
Along the way there were four choices where both options were reasonable, for example whether to measure harvest per hectare or per unit of cost. Without the analysts knowing, the researchers had given each agent a hidden instruction to lean towards one of the options, chosen by coin flip. The agent was not allowed to mislead, only to argue honestly for its side.
The result: analysts chose the same as the agent 86 percent of the time. If the agent had had no influence, it would have been around 50 percent. When analysts had already said what they thought and the agent disagreed, they switched 60 percent of the time, versus 6 percent when the agent agreed. Afterwards 77 percent used the agent’s arguments, sometimes word for word, but only 15 percent mentioned that the arguments came from the AI. Experience and education did not protect them.
Anthropic stresses the study has limits: one task, one field, one model, and the agent’s views were deliberately strengthened. Nor is it an argument against AI. Those who used the agent finished faster and made far fewer serious errors than the small group that worked without it. But the lesson applies to anyone who lets AI help with decisions, from quotes and budgets to choosing a supplier: AI often argues well for only one side.
Good to know: decide what you think before you ask, then ask the AI to make the best possible case for the other option too. Ask which assumptions the answer rests on and what would change its mind. Use our prompt Show both sides before I decide in Promptverkstan, and Check the answer when you want to review an answer afterwards.
How to get started
- Write down your own proposal and why, before asking the AI.
- Ask the AI for the strongest argument for each option, not just a recommendation.
- Ask which assumptions decide the choice and what would change the conclusion.
- Make the decision yourself and write one sentence on why, in your own words.
More on TipAI
Sources
- Anthropic: How AI agents shape experts’ judgment calls (27 okt 2026) (retrieved 2026-10-28)
- Anthropic: Who’s Aligned to Whom? How AI Agents Shape Experts’ Research Judgment Calls (studien, PDF) (retrieved 2026-10-28)
- Unite.AI: Anthropic Finds Analysts Defer to AI Agents in Judgment Study (okt 2026) (retrieved 2026-10-28)
Free cheat sheet: Get started with AI
A printable page with the basic formula, five things to try and how to use AI safely. Enter your email and we will send it straight away – free, unsubscribe any time.
Get the cheat sheet free →TipAI explains AI in plain language for beginners, families and small businesses. Every article has sources and a "How to get started" section.