A new AI benchmark tests whether chatbots protect human well-being techcrunch.com