AI chatbots have failed people in crisis. Can that be fixed?

Clinicians and researchers say AI companies need to open up their safety data.