Scientists asked Bing Copilot - Microsoft's search engine and chatbot - questions about commonly prescribed drugs. In terms of potential harm to patients, 42% of AI answers were considered to lead to moderate or mild harm, and 22% to death or severe harm.