Content Policy
The written rules about what the system may produce.
The documented boundaries a product enforces — what is disallowed, what requires disclaimers, what is escalated to a human. The specification moderation implements.
A written policy makes enforcement arguable and reviewable. Without one, moderation is a black box and every edge case becomes a debate with no reference point.
The specification is the human artifact and the moderation is the implementation, and the two drift apart unless the policy is written precisely enough to be enforced. Vague policy produces inconsistent moderation, which produces user complaints that are impossible to resolve.
A policy a reviewer can apply
Two reviewers classify the same borderline content differently.
The policy says "harmful content" without examples. Rewriting it around the decision — what is allowed, what needs a disclaimer, what escalates — makes the classifications agree and makes the moderation rules implementable.
- standardAI Risk Management Framework: Generative AI Profile (NIST AI 600-1) ↗
The official record for the cross-sector GenAI companion to the AI RMF. Use it as the governance vocabulary; the PDF itself was not machine-readable in this research pass, so do not quote definitions from here.
- reportAnnouncing our updated Responsible Scaling Policy ↗
Capability thresholds with graduated safeguards and documentation duties, written in the safety-case style: here is the evidence that this risk is adequately handled.
Moderation
Filtering inputs and outputs against a policy.
Guardrails
Constraints on what the system will do, enforced in code.
Model Card
The documentation shipped with a model.
Data Governance
Rules about what data may be used, by whom, and retained how long.