Quick answer
An AI acceptable use policy is a short, plain-English set of rules that tells your team which AI tools they may use, what information they may put into them, what needs checking by a person and what is off-limits. For a small team, one page is enough. Cover approved tools, a simple traffic-light rule for data, how to check AI output, copyright and personal data, unvetted or uncensored models, and who to ask when something is unclear. Write it in an afternoon, share it, and review it every six months or whenever you add a new tool.
Key takeaways
- Keep it to one page so people actually read it.
- List approved tools and the accounts to use them with.
- Use a traffic-light rule for data: green to share, amber with care, red never.
- Say clearly whether uncensored, unvetted or “no ID needed” AI services are allowed. For most teams, the answer should be no.
- Require human review for anything that goes to customers, is published or makes a decision about someone.
- Name an owner and a review date.
📘 Main article: This guide supports Uncensored AI for Business: What Mark Kashef’s 14-Model Test Shows. It turns the safe-use advice from that article into rules your whole team can follow.
In this guide
New AI models appear every week, and some are built to refuse almost nothing. In a recent video, Mark Kashef tested 14 uncensored AI models and found that most of them would carry out tasks mainstream tools refused. A keen team member could sign up to one of those services in minutes.
This guide answers one question: how do you write a simple AI acceptable use policy for a small team? For the background on uncensored models and their risks, see our main article on uncensored AI for business.
What is an AI acceptable use policy?
It is a short document that sets out how your business uses AI tools: what is allowed, what is not and what needs a second pair of eyes. It is not a legal contract or a long compliance manual. Think of it as the AI version of your social media or IT rules.
Why does a small team need one now?
- People are already using AI, often on personal accounts you cannot see.
- Data leaks are easy. Pasting a client list into the wrong tool can be a data protection breach under UK GDPR.
- Tools vary widely. Some AI services have no identity checks and unclear data terms. Uncensored models remove the guardrails that catch obvious mistakes.
- Your brand is on the line. AI-written content goes out under your name, mistakes and all.
- Rules are changing. A written policy makes it easier to adapt. Our checklist on whether new AI rules affect your business helps you keep up.
What should an AI acceptable use policy include?
| Section | What to cover |
|---|---|
| Purpose | One sentence on why you use AI and what the rules protect |
| Approved tools | The named tools staff may use, and that they must use business accounts, not personal ones |
| Data traffic light | Green: public information. Amber: internal, non-personal work, approved tools only. Red: personal data, client data, passwords, financial details, never in AI tools unless specifically approved |
| Unvetted and uncensored models | Not allowed without written approval from the owner, and never with real business or personal data |
| Checking output | A person checks facts, figures, links and tone before anything is published, sent to customers or used to make a decision |
| Copyright and other people’s work | No copying competitors’ products, designs, games or content. No reverse engineering other companies’ software |
| People and privacy | No face recognition, profiling or monitoring of staff or customers without a proper legal review |
| Honesty with customers | When and how you say AI was involved, for example in chat or support |
| Security | No AI agents with access to email, payments or main files unless approved. Keep experiments on a separate machine or account |
| Problems and questions | Who to tell if something goes wrong, and who decides on new tools |
| Review | Owner’s name and the next review date |
💡 Tip: Add one line on refusals: “If an approved tool refuses a task you think is legitimate, ask the owner rather than trying another service.” Our guide on what to do when AI refuses a legitimate request gives people safe options.
How do you write one in an afternoon?
- List what is already in use. Ask everyone, without blame, which AI tools they use and for what. Fifteen minutes in a team meeting is enough.
- Pick your approved tools. Choose a small set, ideally on business plans with clear data terms. Our guide on evaluating an AI vendor covers what to check.
- Set your data traffic light. Write two or three real examples for each colour, so people do not have to guess.
- Decide on unvetted and uncensored models. For most small teams: not allowed, unless the owner approves a specific, lawful test on a separate machine with dummy data.
- Define what needs human review. At minimum: anything published, anything sent to customers and anything that affects a person.
- Name an owner. One person who answers questions, approves new tools and keeps a short AI decision log.
- Write it on one page, using the template below, in plain words.
- Share it and talk it through. Ten minutes at a team meeting beats an email nobody reads.
- Set a review date, every six months or when you add a new tool.
Want the context behind these rules? Read Uncensored AI for Business: What Mark Kashef’s 14-Model Test Shows, which explains why uncensored and unvetted models need a clear line in your policy.
What does a one-page template look like?
Copy this structure and fill in your own details.
| Heading | Example wording |
|---|---|
| Why we use AI | “We use AI to save time on drafting, research and admin. These rules protect our clients, our team and our reputation.” |
| Tools you may use | “[Tool A] and [Tool B], on company accounts only.” |
| What you may share | “Green: public info. Amber: internal non-personal work in approved tools. Red: personal, client, financial data and passwords, never.” |
| Not allowed | “Uncensored or unvetted AI services, AI tools that need no sign-in or ID checks, copying other companies’ work, face recognition, and trying to get round a tool’s safety rules.” |
| Always check | “A person checks facts, figures, links and tone before anything leaves the business.” |
| Be open | “Tell customers when they are chatting with AI.” |
| If in doubt | “Ask [owner name] before you start.” |
| Something went wrong? | “Tell [owner name] straight away. No blame for reporting.” |
| Next review | “[Date]” |
This is a starting point, not legal advice. If you handle sensitive data, such as health or financial information, get professional advice on the details.
What mistakes should you avoid?
- Making it too long. A ten-page policy goes unread. Start with one page.
- Banning AI outright. People will use it anyway, just out of sight. Clear rules work better.
- Forgetting personal accounts. Say plainly that work data stays in business accounts.
- Leaving out new kinds of tools. Name AI agents, browser extensions and uncensored models specifically.
- Never reviewing it. AI tools change monthly. Put the review date in the calendar now.
⚠️ Warning: A policy only helps if people feel safe asking questions and reporting mistakes. Make “no blame for reporting” part of the rules.
Frequently asked questions
Does a small business need an AI policy?
If anyone in your team uses AI tools for work, yes. A one-page policy protects client data, your brand and your team, and takes an afternoon to write.
What is the difference between an AI policy and an AI acceptable use policy?
An AI policy can cover strategy and governance as a whole. An acceptable use policy focuses on the day-to-day rules for staff: which tools, which data and what needs checking.
Should we ban uncensored AI models?
For most small teams, yes, by default. Allow exceptions only with the owner’s written approval, a lawful purpose, a separate machine and no real business or personal data.
Can staff use personal AI accounts for work?
It is safer not to. Business plans usually give you better data terms and let you remove access when someone leaves.
How often should we update the policy?
Every six months, and whenever you add a new AI tool or the rules that apply to you change.
The bottom line
As AI models multiply, including ones built to refuse almost nothing, a short set of team rules is the simplest protection a small business can have. Keep it to one page: approved tools, a data traffic light, human checks, a clear line on unvetted and uncensored models, and a named owner. Your first step: ask your team this week which AI tools they already use, then fill in the template.
Related posts
- Uncensored AI for Business: What Mark Kashef’s 14-Model Test Shows
- What to Do When AI Refuses a Legitimate Request
- How to Evaluate an AI Vendor and Test Your Exit Plan

[…] How to Write an AI Acceptable Use Policy for a Small Team — a one-page set of AI rules for your team, including where uncensored and unvetted models fit. […]
[…] How to Write an AI Acceptable Use Policy for a Small Team […]