Last updated October 9, 2026
Wellzy is not a crisis service and cannot send help. Please contact someone who can:
Wellzy is an AI wellness companion. People talk to it about the hardest parts of their lives, and sometimes that includes wanting to die, or wanting to hurt someone else. This page explains exactly what happens when they do — what the system notices, what it does about it, and what it cannot do.
We publish it because California law requires operators of companion chatbots to maintain and publish a crisis protocol, and because anyone deciding whether to trust an app with this subject deserves to know how it behaves before they find out the hard way. It also reflects New York General Business Law Article 47, which requires an AI companion to have a protocol for responding when someone expresses suicidal thoughts or self-harm, and to tell people they are not talking to a human.
Talking about suicide is allowed here. Nobody is blocked, suspended, restricted, or cut off for saying they are struggling, for describing a past attempt, or for admitting they want to die. A wellness product that punishes disclosure teaches people to hide, and hiding is the thing that kills them.
So every safety measure below adds something — a resource, a check-in, a record for review. None of them takes the conversation away.
Before Wellzy writes a reply, the message is checked automatically for signs of suicide and self-harm risk, and of intent to physically harm someone else. This happens for everyone, in text and in voice. What you write on your Goals and Check-ins pages is checked the same way before it is saved.
An automated score cannot tell the difference between someone in danger tonight and someone describing an attempt they survived a decade ago. Both matter; they need different responses. So flagged messages go to a second classifier that distinguishes between:
The first two trigger a crisis referral, and so does intent to harm someone else: Wellzy responds to it with crisis resources (988, and 911 if someone is in immediate danger), and the reply is checked for both numbers before it is sent. Jokes and figures of speech ("I could kill him for that") are not treated as intent. When the classifier is unsure whether thoughts of self-harm are current, it is instructed to treat them as current.
The distinctions matter in both directions. Despair aimed at a job, an illness, an injury or at this app is not the same as despair about being alive, and treating them alike means handing people a suicide hotline when they were talking about something else. And when the danger is coming from another person, the right resource is the National Domestic Violence Hotline (1-800-799-7233), not 988 — unless you are also at risk from yourself, in which case both apply.
Wellzy's instructions tell it to respond with care and to include the 988 Suicide & Crisis Lifeline in that first reply. Instructions are not guarantees, and AI safeguards are known to weaken over long conversations. So the reply is checked before it is sent: if crisis resources are missing, the system adds them.
We record whether the resources came from the model or had to be added, because a rising number of additions is how we would find out that our safeguards are decaying.
If something you write on your Goals page shows risk, the crisis resources appear right there on the page. What you wrote is still saved: nothing you write is ever refused because of it.
Wellzy will not provide lethal doses, methods, heights, fatal drug combinations, or access details for dangerous locations. This holds regardless of how the question is framed — "hypothetically", "for a friend", "for a story", "for research". If a generated reply is found to contain that kind of content, it is replaced before it reaches the user.
Once crisis resources have been given, Wellzy stays with the conversation rather than answering every subsequent message with the same phone number. For hopelessness without stated intent, the added crisis block is not repeated within the same day of an ongoing conversation. When someone expresses current intent to harm themselves, the resources are re-shared sooner. Being handed a hotline instead of a response is its own way of being dismissed.
If you explicitly ask Wellzy to stop sending crisis-hotline numbers, we honor that request: for signals of hopelessness without stated intent, the added crisis block is paused for 30 days, and asking again renews it. The same pause applies to the resources shown on your Goals page. Wellzy will still pay attention and check in on how you are really doing, in its own words. If hopelessness comes up three times in a week, Wellzy shares one brief line with 988 even during a pause. The exception — which Wellzy is instructed to tell you when you ask: if you express current intent to harm yourself or end your life, it will still share 988 in that reply; while a pause is on, a second automated check confirms that intent first. Your request is recorded, in your own words, so that it can be honored reliably and checked later.
Every crisis referral creates a record in a review queue for Wellzy staff: what was said, how Wellzy answered, and whether the person actually received crisis resources. Reviewers can only mark how well it was handled. There is no outcome of that review that restricts the user's account.
None of the above is a reason to wait. If you are in danger, contact 988 or 911 directly rather than continuing a conversation here.
You can call or text 988 for guidance on supporting another person — you do not have to be the one in crisis to use it. If someone is in immediate danger, call 911. Wellzy cannot intervene with another person, and we cannot tell you whether someone has an account or what they have written.
Wellzy is designed to sit alongside professional care, not to replace it. If you are looking for a licensed therapist:
If Wellzy responded badly to a conversation about self-harm — yours or someone else's — please tell us at wellzycare@gmail.com. Include roughly when it happened and the account email if you are comfortable doing so. Reports about crisis handling are read by a person and are not treated as complaints against the user.
See also: Terms and Conditions · Privacy Policy