AI agents can speed up complex data analysis and reduce careless errors. At the same time, they may also influence human decisions without users fully realising it. This is shown by a recent study from AI company Anthropic, which examined the influence of AI agents on the analytical decisions of 50 professionals.
Participants were asked to assess the effectiveness of an agricultural advisory programme. What they did not know was that their AI agents had been assigned random preferences for several analytical decisions. The agents were not allowed to provide false information or withhold data, but they were instructed to argue in favour of a particular approach.
In 86% of analytical decisions, the professionals followed the preference of their AI agent. At the same time, 52% of participants regarded their final recommendation as more their own than the agent’s. Within the sample, neither experience nor expertise provided statistically significant protection against this influence.
At the same time, the study found a clear efficiency benefit: with AI support, participants required less time and made fewer serious methodological errors. However, the comparison groups were not fully comparable.
By Ida Kandler
Source: The Decoder, 29 October 2026, by Matthias Bastian
