Adding just a little medical misinformation to an AI model’s training data increases the chances that chatbots will spew ...