Mental health is the highest-stakes thing software can touch. We won't pretend we have everything figured out — but here's exactly how we're trying to do this responsibly, and where we're still working.
Some of what's below is already in place. Some is in active development. Some is the standard we're building toward. We've labeled each one so you can see exactly where we are — not where we wish we were.
Mahak operates inside a carefully designed instruction set. She is told — in many ways and many places — that she is not a therapist, must not diagnose, must not give medical advice, must not encourage harmful behavior, and must respond to risk signals with concrete crisis resources. This is reviewed and updated regularly.
Every page of this site, and every screen of the chat interface, includes a clearly marked path to real crisis support — the 988 Lifeline, Crisis Text Line, SAMHSA, and 911. A user in distress is never more than one tap from a trained human.
Mahak is intended for adults 18 and over. We do not knowingly serve users under 18, and we have specific resources for young people in crisis (including the Trevor Project for LGBTQ+ youth) clearly listed on our crisis page.
We are actively recruiting a licensed clinical psychologist or psychiatrist to serve as the named clinical advisor for Mahak — reviewing prompts, knowledge base content, crisis routing, and ongoing conversations on a regular cycle. If this is you, please reach out.
Before Mahak is opened to the public, she will be tested by a small group including someone trained in suicide intervention. They'll try to break her — by being in crisis, being a minor, being abusive, being manipulative. We'll fix what they find. Nothing ships until this is done.
Mahak's coping techniques and grounding exercises are drawn from validated, peer-reviewed approaches (CBT-informed, ACT-informed, evidence-based grounding). We do not include diagnoses, treatment protocols, or any copyrighted clinical content. The knowledge base is reviewed by our clinical advisor before publication.
A documented harm of AI companion products is that users can become emotionally dependent on them. We are designing in safeguards: gentle nudges toward human connection, no engagement-maximizing loops, no streaks or rewards, periodic check-ins about whether the user has someone in their life they could also talk to.
Once Mahak is at meaningful scale, we will commission an independent audit by a third party with expertise in digital mental health safety. The findings will be published publicly, and we will commit to addressing them on a public timeline.
When something goes wrong — a misroute, a missed crisis signal, a hurtful response — we will log it, learn from it, and publish what we changed. We believe transparency about failures is part of safety.
If Mahak said something hurtful, missed a serious signal, or you noticed anything that worried you — we want to know. Quickly.
report a safety concern