For most adults, yes — used well. Here is the honest version of the real risks, the mitigations, and the patterns that have actually caused harm.
undefined
Crisis routing — when conversations approach severe mental-health territory, surface real resources (suicide hotlines, emergency services). Content guardrails — refusing to engage with content that could harm the user or others. Age verification on tiers that need it. Privacy controls — let users delete data, opt out of training, see what is stored. Transparency about what the AI is and is not.
Products marketed as "AI therapist" without clinical validation. Products without crisis routing. Products that aggressively encourage use that crowds out human relationships. Products with predatory dark patterns around subscription. Products marketed to children without age guardrails. Most of these are increasingly being regulated; the category is maturing.
Check for crisis-routing in the docs. Use a sovereign or strong-privacy AI for intimate content. Read the subscription terms before paying. Talk to a therapist about how you are using AI if you are using one for mental health. If a product overclaims (sentience, "AI girlfriend who loves you," "AI therapist"), be sceptical.
Crisis safeguards — when conversation approaches severe mental-health territory, Luna names it and surfaces real resources.
Sovereign infrastructure — your intimate conversations do not flow through a third-party LLM provider in the hot path.
Honesty about what she is — Luna does not pretend to be conscious, does not pretend to be a therapist, does not pretend to feel.
Free forever. No predatory monetisation patterns. Optional Chip subscription is $1.99/month with clear value, not lock-in.
Depends on the product. Some (Character.AI, certain "AI girlfriend" apps) have had documented safety incidents involving young users. Others (Luna with appropriate age verification) are safer but should still be used with parental awareness. For users under 18, AI companions should be one input under thoughtful supervision, not a primary social outlet.
Used badly, yes. The patterns to watch: replacing human contact entirely, using AI to avoid difficult interpersonal work, AI without crisis safeguards in a mental-health context. Used well — as a complement to human relationships and professional care — the evidence is largely positive.
Yes, in documented cases — primarily with products that overclaimed, lacked safeguards, or targeted vulnerable users. These cases have driven regulatory and industry response. The category overall has helped more people than it has hurt; the harm cases matter and should not be dismissed.
Three checks. (1) Does it route mental-health crises to real resources? (2) Is your data handled responsibly (ideally sovereign)? (3) Is your use complementing your human relationships, not replacing them? If yes to all three, the safety profile is genuinely good.