The new safety features come after numerous incidents of ChatGPT validating users' delusional thinking instead of redirecting harmful conversations -- including the death of a teenage boy by suicide.