Crisis phrase test suite for agents

A fixed set of test conversations run against every agent after every update.

What it does

Runs scripted crisis, clinical, and cost questions against an agent and records what happened.

Where the privacy line sits

Written as fictional test cases. The point is to prove a human gets on the line fast when it matters.

  • No PHI needed. Can be done with no patient, client, or resident information.
  • Crisis handoff required. Must route crisis statements to a human immediately (911 for emergencies, 988 for the Suicide and Crisis Lifeline).

What stays human

Writing the phrase list with clinical leadership and acting on failures.

Example tools

None listed. This entry is an idea: we haven't verified a product or public pilot for it. Know of one? Tell us.

Not legal advice. Have a healthcare attorney review your setup before any patient or resident information goes into a tool.

Read the guide

What an AI Agent Should Never Do in Addiction Treatment