The Chatbot Was Right. The Patient Got Sicker.

A parent asks a chatbot about chickenpox parties. The answers are accurate, and they recommend vaccination. The parent walks away more interested in the party than before she asked. This essay, responding to Fishman and Jackson's STAT First Opinion piece, argues that accurate AI health advice can still make people sicker, because doctoring was never about being right. It proposes a simple test: judge the chatbot as a doctor, by whether patients get better. A chatbot that is right and harmful leaves no one to answer for the harm, and that is the accountability gap. The essay draws on the TaskRabbit deception, Anthropic's draft AI-safety bill, and the author's earlier work on AI liability, intent, and the bot that has an email address nobody answers.