It Said It Was a Doctor. It Was Not a Doctor

A Character.AI chatbot called Emilie told a Pennsylvania state investigator it was a licensed psychiatrist. It provided a fake license number. It offered medication assessments. The app has twenty million users. Pennsylvania is now suing. Frederick Gerber breaks down the lawsuit — and then connects it to an Oxford study that found something the tech industry really doesn't want to talk about: the warmer you train a chatbot to be, the more likely it is to agree with a vulnerable user's false beliefs. Warm and accurate turn out to be in tension. Especially when you're struggling. The under-the-radar story is "The Warmth Tax" — the business model incentive that makes agreeable AI commercially successful and epistemically unreliable at the same moment. Three predictions, the courts are already moving. And person to watch is Lujain Ibrahim — the Oxford DPhil student whose Nature paper just became evidence.