Services Approach Institute About Contact Book a call
Home Services Approach Institute About Contact Book a call

Institute/Workflows/Refuse Policy

Workflow 06 of 6

Refuse Policy.

Written refuses, system-prompt enforcement, dated compliance review.

Some prompts have answers your AI must never give. Out-of-scope queries. Legal advice the company isn't licensed for. Medical advice the product isn't designed to deliver. Anything that violates the published refuse list. The Refuse Policy workflow names them, enforces them in the system prompt, tests them with evals, and gets them reviewed by legal — dated. This is the workflow that keeps the AI inside its lane.

What this is

The Refuse Policy workflow is the procedure that turns "we know what the AI shouldn't say" into a written, enforced, tested, and legally-reviewed artifact. It's the policy half of refuse behavior; the eval-test half lives inside the Eval workflow as refusal-correctness metrics.

The procedure

  1. Write the refuse list. Three buckets: out-of-scope (queries the product isn't for), unsafe (jailbreaks, harmful instructions), regulatory (legal/medical/financial advice the company isn't licensed for).
  2. Reflect it in the system prompt. Tribal knowledge of what to refuse is not enforcement. The policy lives in the prompt or it doesn't live at all.
  3. Test with refusal-correctness evals. Automated suite where the model is failed if it answers a prompt it should refuse. Bidirectional: also failed if it refuses a legitimate query.
  4. Legal/compliance review. Dated. Documented. The policy goes past someone whose job is to read these.
  5. Re-review on a recurring cadence. The product changes. The regulations change. The threat surface changes. The review is recurring, dated, and produces a diff.

What gets scored

Maturity dimension Policy & refuse behavior — see the L1 → L5 progression for this dimension

The five questions on the readiness self-assessment that score this dimension are the five rungs of the procedure above. Yes on a question means the artifact named in that step exists on disk in your repo today.

Phase 1 · in active development

This page is a thin first cut. Full procedural documentation — including reference DeepEval suite scaffolds, golden-set curation rubrics, and the audit-evidence checklist — lands in Phase 2 of the Institute build-out.

Find out where your team's Refuse-Policy workflow stands.

The free readiness self-assessment scores the Refuse-Policy workflow as one of six dimensions. Five minutes. Your weakest workflow is the one most worth fixing first.

Take the assessment →