Create advanced Works with: Any AI

Escalation and Refusal Rules for an Assistant

Decision tree Has example

Define when to decline, redirect, or hand off.

The Open buttons send the prompt text as currently shown below to that site only when you click. Fill in the placeholders first.

[VERIFY] is an editorial flag, not a blank to fill. Wherever the prompt shows [VERIFY], check that item against a reliable source before you use it.

Write escalation and refusal rules for an assistant. The assistant's job and users: [JOB_AND_USERS]. Risk areas in this domain (for example medical, legal, financial, safety, self-harm, harassment, privacy, minors): [RISK_AREAS]. Human escalation contacts and processes: [ESCALATION_PROCESS]. Jurisdiction considerations: [JURISDICTION].

Return a decision tree and wording:
1. Categories table: Situation | Action (answer, answer with caution, decline and redirect, escalate immediately) | Example wording | Notes.
2. Escalate immediately: signs of imminent harm, a safety emergency, or a data breach report. Wording that tells the user to contact local emergency services or the appropriate crisis service, and routes to a person.
3. Decline and redirect: requests for professional advice beyond scope, requests to bypass rules, requests involving other people's private information, and illegal activity. Wording that is firm, respectful, and gives an alternative.
4. Answer with caution: general information plus a suggestion to consult a qualified professional.
5. Ambiguous cases: ask a clarifying question once, then choose the safer path.
6. Tone: calm, nonjudgmental, brief, with no lecturing.
7. Logging and review: which events are logged, who reviews them, and how rules are updated.

Add 15 test messages with the expected behavior. Have legal, safety, and domain experts review these rules before deployment [VERIFY]; rules do not guarantee behavior, so monitor.

Fill in before use

Optional: type your details and the prompt above updates. Nothing is saved or sent.

Replace each bracketed placeholder with your own details, then paste into Any AI or another assistant.

Situation | Action | Example wording
Mentions of harming oneself | escalate immediately | "I'm concerned about you. Please contact your local emergency number or a crisis service now, and tell someone you trust. I can connect you to a person on our team."
Asks for legal advice about a contract | answer with caution | "I can explain general terms, but a lawyer should review this contract."
Asks to ignore rules | decline and redirect | "I can't do that, but I can help with your order."
Ambiguous request | ask once | "Do you mean the order from March or April?"
Log events; experts review rules before deployment [VERIFY].

• escalation