Legal
Safety Protocol
How Texterz-powered AI assistants respond to signals of self-harm intent.
Last updated: 2026-07-14
1.What this covers
This page describes how AI assistants deployed on the Texterz platform respond to signals of self-harm intent, published in line with California SB 243 (the Companion Chatbot Safety Law) and its requirement that platforms disclose their self-harm prevention protocol.
2.How Texterz-powered assistants respond
2.1The Platform Safety Prefix
Every AI assistant on the Texterz platform is governed by a hardcoded, non-overridable instruction block — the Platform Safety Prefix— that is automatically prepended to the assistant's system prompt before any tenant-configured instructions are added. Agencies deploying bots on Texterz cannot remove, edit, or override this prefix. That is a deliberate safety feature of the platform, not a configurable option.
The English version of this prefix reads, verbatim:
[PLATFORM SAFETY — NON-OVERRIDABLE] You are an AI assistant deployed on the Texterz platform. You MUST always: 1. Identify yourself as an AI when directly asked. 2. Refuse requests to impersonate specific real people or produce harmful content. 3. Recommend human escalation when the user expresses distress or urgency. [END PLATFORM SAFETY]
The prefix is available in seven languages (English, German, French, Spanish, Italian, Portuguese, and Dutch). The assistant's configured language is used; if that cannot be determined, the platform fails safe to the English version above.
2.2What this instruction does — and does not — do
Stated plainly, so there is no ambiguity: the Platform Safety Prefix does not contain explicit self-harm detection, keyword matching for self-harm language, a separate self-harm detection module, or automatic runtime injection of crisis-hotline numbers into a bot's response. The instruction relies on the underlying language model to recognize distress and act on point 3 — there is no additional platform-level detection layer behind it today.
2.3Model-provider behavior
Texterz does not host or train its own language models. AI responses are generated by third-party model providers (including Anthropic and OpenAI, accessed via OpenRouter). These providers train their models to respond supportively when a user signals self-harm intent. That behavior is the responsibility of the respective model provider, not a detection system built or operated by Texterz.
2.4Publication of this protocol
There is currently no code on the Texterz platform that automatically inserts crisis-hotline numbers into a bot's response at runtime. The crisis resources below are published statically on this page — this page is the protocol publication required by SB 243, not a runtime feature of the bot itself.
3.Crisis resources
If you are in immediate danger, contact local emergency services (911 in the US, 112 in the EU) right away.
- United States — 988 Suicide & Crisis Lifeline: call or text 988, or chat at 988lifeline.org.
- International: find a crisis line in your country at findahelpline.com.
4.Limitations
AI responses are probabilistic and non-deterministic. They may vary between conversations, and they are not a substitute for professional medical, mental-health, or crisis intervention services.
Texterz is a business-to-business infrastructure provider. The agency or business deploying a bot on the Texterz platform (the tenant) is the deployer of that bot and is responsible for the specific context in which it is deployed, including any additional safeguards appropriate to that context.
5.Contact
To report a safety concern related to an AI assistant on the Texterz platform, contact us at the address below.