Writing a Team AI Policy
Write a short, practical team AI policy: approved tools, what data can go where, when to tell clients, and how to escalate, all within your organization's rules.
Start From What Already Exists
Before writing anything, find out what already applies to your team:
- Your organization’s AI policy, if it has one, and its information security and privacy policies.
- Legal and regulatory requirements for your industry and location. Data protection laws, sector rules (health, finance, education, government), and AI-specific regulations such as the EU AI Act may apply. Ask your legal, compliance, or privacy team. Don’t guess.
- Contracts. Client agreements may restrict how their data can be processed, or require disclosure of AI use.
- Approved vendors. IT or procurement may already have approved AI tools with proper data agreements.
Your team policy fills in the practical details within those rules. It should never contradict them.
Five Things a Team Policy Must Answer
1. Which Tools Can We Use?
List the approved tools and what each is approved for. Be specific: “Our company AI assistant (enterprise account) for all internal work” is clearer than “approved AI tools.” Say how to request a new tool.
2. What Data Can Go Where?
The most useful part of any policy is a simple data classification that maps to tools:
| Class | Examples | AI use |
|---|---|---|
| Public | Published marketing, public website content | Any approved tool |
| Internal | Internal processes, non-sensitive project notes | Approved enterprise tools only |
| Confidential | Client data, financials, contracts, unreleased plans | Only tools specifically approved for confidential data, or de-identify first |
| Restricted | Personal data about individuals (health, HR records), credentials, regulated data | Not in AI tools unless explicitly approved by security, privacy, or legal |
If your organization already classifies data, use its categories. Don’t invent a second system.
3. When Do We Tell Clients and Stakeholders?
Decide disclosure rules up front, based on contracts and on what clients would reasonably expect. For example: “We disclose AI use in client deliverables when AI drafted substantial content or analysis. We never present AI output as a named expert’s personal judgment.”
4. What Must a Human Do?
Point to your workflow maps and review tiers (Modules 1 and 2). Name the work that always stays with people, such as final decisions about people, legal commitments, and pricing.
5. What Do I Do When Something Goes Wrong?
Make escalation simple: who to tell about a data mistake (for example, confidential data pasted into the wrong tool), an AI error that reached a client, or an uncertain case. Make clear that reporting quickly is always the right move, and won’t be punished.
Make It Usable
- Keep it short. One or two pages. People don’t follow what they can’t remember.
- Use examples. “Okay: summarizing our public blog posts. Not okay: pasting a client’s contract into a personal chatbot.”
- Say why. People follow rules they understand.
- Review it regularly, at least twice a year and whenever tools or regulations change.
Common mistake: a ten-page policy full of legal language that nobody on the team reads, so everyone keeps doing whatever they were doing.
Your organization may need a formal policy. Your team also needs a practical one-pager that translates it into everyday decisions, with examples. Ask a team member to explain it back to you. If they can’t, rewrite it.
- Gather your organization’s existing AI, security, and privacy policies, and note any client contract terms about AI or data.
- Draft a one-page policy covering the five questions above, with at least one “okay” and one “not okay” example for each.
- Ask an AI assistant (using an approved tool, with no confidential details) to stress-test it:
Here's a draft team AI policy. Act as a new employee in their first week. List every situation where you'd be unsure what to do, and any rule that seems to conflict with another. - Revise. Then have your legal, compliance, or IT contact review it before rollout.
- Plan how you’ll walk the team through it, and when you’ll review it next.
1. What should you check before writing a team AI policy?
A team policy fills in practical details within existing rules. It must never contradict them.
2. An employee wants to paste a client's contract into an AI tool to summarize it. Under the example data classification, what applies?
Client contracts are confidential. They need specifically approved tools, or the sensitive details removed.
3. What makes escalation rules effective?
Fast reporting limits damage. People report quickly only when it's safe to do so.
Up Next
Some AI uses carry higher stakes than others, especially when decisions affect people. The next lesson covers risk, fairness, and accountability.