The team then assessed how a basket of popular AI models complied with these guidelines. Models tested included Meta’s LLaMA 3 and OpenAI’s GPT-4o, as well as purpose-built therapy chatbots such as ...