A new study finds AI-driven chatbots provide problematic or misleading medical advice approximately half the time, raising serious healthcare safety concerns.