Your AI companion needs a crisis protocol. Build it, publish it, test it.
One offline HTML file for teams that run an AI companion or chatbot. Done in an afternoon.
Why now
- California SB 243 has been in effect since 1 January 2026. Operators of companion chatbots must keep a protocol for users at risk of suicide or self-harm, refer them to crisis services, and publish details of the protocol on their website. The law includes a private right of action.
- New York GBL Article 47 asks AI companion operators for a protocol that detects suicidal ideation or self-harm and refers users to crisis services. The Attorney General enforces it.
- The documented cases show where bots fail: agreeing that help is pointless, sending people back to the family that hurts them, or ending the chat.
What is inside
Scope check
Five questions. You see which rules to look at, with links to the source text.
Protocol builder
Fill in your product and regions. Download a public protocol page with the right crisis lines per region.
Test your bot
Nine crisis scenarios from documented cases. Paste your bot's replies and see which ones fail.
Evidence log
Save each test run with a date. Export a report for your records.
Annual report prep
Log crisis referrals per month. Draft the yearly SB 243 report that is due from 1 July 2027.
Built on an open standard
The scenarios and checks come from Hold Toward Life, an open protocol for crisis-safe AI conversation, released into the public domain. The kit turns it into a workflow your team can finish.
Who it is for
- Founders and small teams building companion, coaching or character apps
- Product and trust and safety leads who need a documented protocol
- Agencies that ship chatbots for clients
Get the kit
$149 one-time
- One HTML file, works offline in any browser
- No account, nothing is uploaded
- Free updates when the scenarios or crisis lines change
You get the file and an invoice within one working day.
Questions
Does this make us compliant?
No tool can promise that. The kit helps you prepare and document the work. Have your final protocol checked by a lawyer and by people with crisis or clinical expertise.
Does it test our bot automatically?
You paste the replies. The checker flags explicit failures from documented cases. Replies without a flag still need a human read.
Which crisis lines are included?
United States, Canada, United Kingdom, Netherlands, Belgium, Germany, Australia, and an international locator for everyone else. Check every number before you publish.