Vidyanand Choudhary, professor of information systems, gave widely used language models a task that sounds almost too simple to be interesting: look at two résumés, and pick the stronger candidate. Then he handed the models the same two résumés again, in the opposite order. A striking number of times, the AI models changed the answer. Choudhary set out to look at AI hiring biases, like gender and ethnicity. Instead, the results revealed AI models are easily swayed by simple layout, frequently changing who they think is the best candidate because a résumé was moved from the top to the bottom.