A study of this sort cannot evaluate the intensity of responses over time, as in Adam Raine's case. I find it unconscionable that chatbots have been promoted as therapeutic agents - they are utterly unregulated, with no provision to monitor users' experiences.
I should clarify that by “monitor,” above, I mean “get feedback when the product is malfunctioning” (eg, reinforcing suicidal plans or giving other unacceptable, life-endangering responses).
(And I agree with Brian that OpenAI needs to be named and blamed specifically.)