yes it's probably true
> "No computer can hope to accurately respond in a way that isn't harmful at least some of the time."
yeah so far it looks like that's the case
> "Human's don't have that problem."
wait what
human's can handle multiple different meanings given the same inputs and provide a correct response.
the computer science concepts of the halting problem (https://en.wikipedia.org/wiki/Halting_problem) or decidability problem (https://en.wikipedia.org/wiki/Decidability_(logic)#Some_unde...) aren't relevant to running an eating disorder helpline
they are when you replace a human's activity with what's effectively a computation engine.
Those have fundamental limitations which are well known, and they did it anyway. Any reasonable professional in a position to advise should have known, there is really no excuse because this is such a fundamental limitation going all the way back to Claude Shannon and Turing. Its discussed regularly in computer science major related undergraduate coursework.
If you are going to use AI, you should be being advised by experts who are knowledgeable in the area especially when there is potential harm as it is safety critical.