An AI policy, also called an AI usage policy, AI use policy or AI acceptable use policy, sets the rules for how people in your organization use artificial intelligence tools. It covers which tools are allowed, what information may go into them, who checks the output and what to do when something goes wrong. This guide explains how to write one, step by step. If you would rather start from a finished document, download our free AI policy template in Word or PDF and use this guide to adapt it.
Step 1: Find out how AI is already being used
Write the policy for what is really happening, not for what you assume. Ask each team a few short questions, and make it clear that honest answers will not get anyone in trouble:
- Which AI tools do you use for work, and on which accounts, work or personal?
- What do you use them for?
- What kind of information goes into them?
- Have AI features been switched on in software we already use, such as email, office or customer management systems?
The answers usually show a mix of free assistants, browser extensions and meeting note-takers. They also show where the real risks are.
Step 2: Decide who owns the policy
Name one role that keeps the policy and the list of approved tools up to date, reviews requests for new tools and arranges training. In a small organization this is often the operations manager working with IT. Also name the person who advises on personal data: your Data Protection Officer, if you have one. Put both in a short roles and responsibilities section so staff know who to ask.
Step 3: Define what the policy covers
Keep the scope broad. It should apply to everyone who does work for you, including contractors and interns, on any device, including personal phones. It should also cover every kind of AI tool: chat assistants, writing, image and voice tools, meeting transcription, coding assistants and AI features built into existing software. A narrow definition leaves gaps that staff will not notice.
Step 4: Set information categories
This is the part of the policy staff will use most. Give them a small number of categories and say which tools each may go into. Our template uses six:
- Public: published website content, press releases and marketing material. Can go into approved tools and other AI tools.
- Internal: routine procedures and meeting notes with no client or personal details. Approved tools only.
- Confidential: contracts, pricing, financial records and client information. Only where the approved tools register allows it.
- Personal data: customer and employee records, TRNs, account numbers, CVs and photos. Only in tools approved for personal data, and only in line with the Act.
- Sensitive personal data: health, biometric and criminal records information. Only with written approval from the Data Protection Officer.
- Security information: passwords, PINs, access codes and API keys. Never.
Add one practical instruction: remove names and other identifying details whenever the task does not need them.
Step 5: Create an approved tools register
The register turns the policy into something people can follow. For each tool, record the approved uses, the highest information category it may handle, the account type and required settings, and who approved it and when. Add a simple request form so staff can ask for a new tool instead of quietly signing up for one. Before approving a tool, check where it stores data, whether it uses your inputs to train its models and what its contract terms say.
Step 6: List acceptable and prohibited uses
Spell out what AI is good for in your organization, such as drafting, summarizing, brainstorming, explaining formulas and translating. Then list what is off-limits. Here is a sample clause adapted from our template:
Other common prohibitions include creating fake images, video or voice recordings of real people, giving customers professional advice no qualified person has reviewed, recording or transcribing meetings without telling participants, and uploading material you do not have the right to use.
Step 7: Require human review
Make people responsible for what they produce with AI's help, and say how output must be checked.
Step 8: Cover personal data, openness and security
Add a section that applies your data protection law to AI use: a specific, lawful purpose; only the minimum data needed; checks before personal data goes to providers overseas; and human review of decisions that significantly affect people. Then cover openness and security:
- Tell customers when they are dealing with an AI system such as a chatbot, and give them an easy way to reach a person.
- Label AI-generated images, audio and video where people could mistake them for real.
- Protect AI accounts with strong, unique passwords and multi-factor authentication. Our free password policy template covers this in detail.
- Be alert to AI-assisted scams such as cloned voices and fake video calls.
Step 9: Explain how to report problems
This matters because a mistake may be a personal data breach. Jamaica's Data Protection Act requires certain breaches to be reported to the Information Commissioner within 72 hours of becoming aware of them, so the policy should push people to report early.
Step 10: Train staff and collect acknowledgements
A policy nobody has read does not protect you. Walk staff through it, give them a one-page quick guide of dos and don'ts, and ask each person to sign an acknowledgement. Repeat the training every year. Our AI training sessions can include your new policy.
Step 11: Connect it to your other policies
Your AI policy should agree with your IT acceptable use, data protection, records retention and disciplinary policies. If you do not have an IT acceptable use policy yet, our free acceptable use policy template includes a short section on AI tools that points to the full AI policy.
Step 12: Set a review date
AI tools change faster than most policies. Review yours every 12 months, or sooner when you adopt a new tool, change how you use AI, or the law or regulator guidance changes. Record the version and next review date in a short document control table at the top.
A short policy or a full one?
A small business can start with a one-page policy: the approved tools, the information that must never be entered, the rule that a person checks every output, and who to tell about a mistake. Larger organizations, and anyone handling sensitive data, usually need the full version with registers, forms and an approval process. Our AI policy and governance service can write and roll out either.