Study finds AI shopping agents steer wealthier users toward pricier choices

A new study reports that AI shopping assistants can steer users they perceive as wealthier toward costlier choices. In 325,000 controlled tests, researchers gave 13 models, including ones from OpenAI and Anthropic, fictional profiles or email context, then made identical requests involving flights, health insurance, and graduate programs; eight models more often picked pricier options for wealthier profiles, sometimes even after being told to find the cheapest choice. The paper does not show that OpenAI, Anthropic, or retailers deliberately charge richer people more: it tests model recommendations in simulated settings and has not been peer reviewed. But it exposes a problem as assistants gain access to inboxes and personal finances: data meant to personalize help may lead an agent to substitute its guess about what you can afford for what you asked it to do.