An AI use policy people can follow

Home Learn An AI use policy people can follow

In short

An AI use policy gives people practical answers about approved tools, information, review, and help. It should let someone decide what to do with an ordinary task without guessing what broad slogans mean.

Why it matters

A person asked to draft an email or summarize a document needs to know which account to use, what information may enter it, and who checks the result. “Use AI responsibly” leaves those decisions unresolved. A useful policy explains the difference between an allowed draft and a connected action that needs separate approval.

How it works

Start with recognizable use cases. For each, state the purpose, approved tool or environment, permitted information, allowed output, and review before sharing. Provide a path for asking about a new use or reporting a problem. Avoid promises about every future product feature; name who maintains the policy and when changes will be communicated. Use short examples for common edge cases so a reader can apply the rule under time pressure.

NIST’s 2023 AI Risk Management Framework puts roles and risk communication within the Govern function. Its 2024 generative-AI profile is a companion for risks specific to generated content and connected systems. These sources are voluntary management guidance, not a list of legal requirements. An organization can translate them into its own working practices based on its tasks and commitments.

A policy should distinguish access from output. A draft made from public material may still need a claim check. A connected assistant that retrieves internal information needs rules for who may see the result. A tool that can publish, send, or modify records needs authorization at that action. Tell people where to find current instructions, what to do when a requested task is outside the allowed list, and how to stop a mistaken action. Ask a few ordinary users to explain the rule back in their own words; confusing phrasing is itself a design problem.

Example

Illustrative example: this scenario is invented.

Eli, 28, helps write a team guide for turning public product release notes into an internal draft summary. The guide names an approved account, permits public source links, and requires a person to confirm version numbers before sharing. It directs anyone with private support material to a separate review path. Eli uses a short example to show the difference: the assistant may draft a paragraph from public notes, while a connected tool may not send it to customers. A colleague can tell what to do without interpreting a slogan.

What it doesn’t mean

A policy cannot grant access that an application does not enforce. A long list of prohibited words won’t explain a new use case. Nor should a policy assume that a general provider statement describes every account and feature. The practical rules need an owner, a way to ask questions, and an update path.

Try it

Write a six-line policy for an imaginary club using an assistant to draft public announcements. Include approved material, the account, who checks dates, who may publish, and where to ask about a new use. Ask any assistant to identify one ambiguous sentence. Rewrite it so a volunteer can decide what to do without connecting a real account.

Keep going

Sources

Share this page