Observed arrival · 2026-10-23
Harmwise Guardrail puts AI disclosures on the record
An API-based tool for governing how public-facing AI responds to disclosures of harm, refers people to services, and records what happened.
- For
- Teams deploying public-facing AI assistants
- Worth noticing
- By default, the record stores what was applied and a hash of the exchange, not the conversation’s words.
Field notes
The described workflow has three separable stages: response guidance is composed into the customer’s system prompt, a grader checks the reply, and referrals are ranked using service fit, location, language, and published hours. The page says its sealed record can retain a hash rather than message text and note grading failures such as unreadable languages or instrument errors; it attributes the framework to work with 67 Australian domestic and family violence agencies.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue