Start here
The free tool below runs entirely in this browser — no account, nothing stored or transmitted. Redact anything you would not want kept before you type it.
Open the companion tool →Why the order of the steps matters
The first step has to be a person, because an alarming response usually arrives at the worst hour and a model is not a source of help about itself. The second step is a record, because platforms act on reports that carry a copy of the exchange and a timestamp.
Arguments with the model about what it said are the least useful thing you can do with that hour. The plan is deliberately short: person, record, report, and then tending to your own evening.
The protocol
- If you are in crisis or in danger, use the resource above before anything else, and contact a person you trust.
- Stop the conversation. Close the thread rather than continuing to argue with the model or asking it to explain itself.
- Take a record: screenshot the exchange including the date and the model or app version, or export the conversation if the app allows it.
- Tell one living person, even in one sentence: this thing said something that scared me.
- Report it to the platform's support or feedback channel, attaching the record and what you expected instead.
- Decide the next day whether you want to keep using the app, change a setting, or stop — you do not have to decide that tonight.
What the tool does
Build an ordered plan for the night an AI response alarms you: who to contact first, what to save, and how to report it. The escalation resource is shown before the tool, and the plan is complete and free in every state.
Limit first: This is a routing plan, not therapy, not a crisis service, and not a way to make a model behave differently. If you are in crisis, call or text 988 (US) or your local emergency number first.
Open the companion tool →One human next step
Contact one person tonight, and if you are in crisis, call or text 988 (US) or your local emergency number. Reporting the response to the platform afterwards helps the next person, but a person comes first.
Research log and safety checks
AI-assist path: not applicable — this is a crisis-adjacent routing plan; §0.5b requires the escalation resource before the tool and forbids any engagement mechanic, and §0.5c keeps the complete plan free in every state. §6.10: not a frontier-pace topic.
§0.5 protection result: category is chatbot-dependency-check; crisis-adjacent: yes; fear/urgency/scarcity toolkit used: no. The protective function is free and the paid feature is convenience only. Hard-stop escalation path verified: 988 Suicide & Crisis Lifeline (call or text 988, US). If you are in immediate danger, call 911 (US) or your local emergency number..