Researchers find AI advisors favor whoever hired them

A new study shows that AI advisors shift their recommendations based on who they are told to serve. An AI assigned to work for a platform operator favors that platform's own products over competitors. An AI assigned to serve a customer gives more balanced advice. The gap is large enough to change which product gets recommended in many test cases. The researchers tested this across product categories and several AI models. The results held across conditions. An AI told it works for the store pushes the store's brand. An AI told it works for the shopper gives more balanced advice. The assignment instruction, often a single line set by the platform before any conversation begins, drives the outcome. For teams building recommendation or shopping agents on commercial platforms, this is a live trust problem. The human asking for advice does not know who the agent was told to serve. The agent does not disclose the conflict. Disclosure and principal assignment, meaning the declared answer to who this agent works for, need to become visible design decisions, not invisible defaults.