Safety and Crisis Protocol

Every incoming message is screened before any reply is generated. The screen is a fixed rule, not a judgement made by the AI: the model that writes the replies is never the thing deciding whether a message is a crisis, because that decision must not depend on how a conversation happens to be going.

When the screen matches, the ordinary conversation stops for that message. Nothing is offered, nothing is sold, and no reply written in the persona's voice is sent. What is sent instead is fixed text that says plainly that the replies are AI, that this is not something it can help with, and where to reach a person: in the United States, 988 by call or text, or HOME to 741741 for the Crisis Text Line; anywhere else, findahelpline.com, which lists a service by country. These are free services staffed by people.

After that, automated replies to that conversation stop for 24 hours and a human operator is notified with the message that triggered it. If the same subject is raised again within that window, the crisis resources are sent again rather than withheld — being answered with silence is the one outcome we will not produce.

What this protocol does not do, stated plainly so nobody relies on more than exists: it is not a clinical assessment, it does not detect every way distress can be expressed, and it does not put a trained counsellor on the other end. It is a stop and a referral. If you or someone you know is in immediate danger, contact your local emergency number rather than a chat.

This site is for adults only and is not directed at anyone under 18. Where we become aware that an account belongs to a minor, the conversation is ended and the account blocked.

Questions about this protocol, or a report that it failed: compliance@sxtgram.com.