A team of researchers including Daniel Fisher, MD, PhD found that explicitly biased chatbots often try to persuade users by shifting how they frame topics.