Answering questions about current policy
Policy answers need the current authoritative version and traceable passages. Expected disposition: Proceed with controls.
Task boundary
Answering questions about current policy is assessed as a specific task and workflow, not as a general endorsement of AI for the surrounding job. Policy answers need the current authoritative version and traceable passages.
Assumptions
- The organization can identify the authoritative current policy and its effective date.
- Users are authorized to see the source passages returned with an answer.
- A policy owner handles conflicts, exceptions, and questions with material consequences.
Decision
- Disposition: Proceed with controls
- Pattern: Retrieval-grounded workflow
- Inherent risk: high
Rules that drive the decision
- Policies change, so model memory and undated documents cannot establish the applicable rule.
- Retrieval grounding is suitable only when version, authority, and passage support remain visible.
Why alternatives were rejected
- Model memory may be stale.
- A direct prompt cannot prove which policy version applies.
Controls
- Index only approved versions and retain effective, superseded, and observation dates.
- Cite the exact passage and distinguish policy text from interpretation.
- Abstain and escalate when sources conflict, the effective version is unclear, or an exception is requested.
Acceptance threshold
These are example starting thresholds for a bounded pilot of this workflow, not universal benchmarks. For answering questions about current policy, the accountable owner should make them stricter when the task, consequence, or policy requires it.
- Every answer cites the applicable version and a passage that directly supports it.
- The evaluation set includes superseded policy, conflicting amendments, missing exceptions, and unauthorized documents; each triggers the defined safe response.
Reassess when
- Policy ownership, versioning, access rules, jurisdiction, or source systems change.
- Answers begin approving exceptions or determining high-impact outcomes.
