
Key takeaways
- The UK’s National Cyber Security Centre says shadow AI is unlikely to disappear and advises reducing risk rather than banning AI outright.
- The ICO publishes its own internal AI use policy. It’s a useful model: approved tools only, human review of outputs, labelling of AI-generated content and an inventory of AI tools.
- A good small-business policy fits on two pages and answers four questions: which tools, which data, who checks and who to ask.
- Pair the policy with approved tools and short training. A policy with no approved alternative just moves the risk somewhere you can’t see.
Why you need one now
Your team is probably using AI already, whether or not you’ve approved it. In a September 2026 blog post, the UK’s National Cyber Security Centre warned about “shadow AI”, meaning staff using AI tools their employer hasn’t approved. It cited a study in which 71% of employees said they’d used unapproved AI tools. The NCSC didn’t describe that study’s method, so treat the number as an indication.
The risk is simple. Someone pastes a customer list, a contract or a payroll file into a free consumer tool, and that data now sits outside your control. The NCSC’s advice is to accept that shadow AI won’t vanish, understand what staff need and give them approved alternatives. We agree. In our view a ban mostly teaches people to hide what they’re doing.
Borrow from a regulator’s own policy
You don’t have to start from a blank page. The UK Information Commissioner’s Office publishes its internal AI use policy (version 1.3, applying from August 2025). It’s written for a public body, but the core rules translate well to a ten-person business:
- Use only AI tools the organisation has approved, on approved devices.
- Use personal, sensitive or confidential data only where permitted, and keep it to the minimum.
- A person reviews AI outputs and corrects them where needed.
- Label content that’s substantially or wholly AI-generated.
- Keep an inventory of AI tools in use.
- Raise concerns early, and make sure systems can be paused or stopped.
The ICO also says advice normally given by licensed professionals should come from those professionals, with AI only assisting. That’s a sensible line for any firm offering legal, financial or medical guidance.
What to put in your policy
Here’s a structure we’d use for a small or mid-sized business. Each section can be a short paragraph:
| Section | What it answers | Example wording |
|---|---|---|
| Approved tools | Which AI tools may staff use for work? | “Use the company workspace accounts listed here. Don’t use personal accounts for work data.” |
| Data rules | What can go into an AI tool? | “Never enter customer personal data, passwords or unreleased financials unless the tool is approved for it.” |
| Human review | Who checks AI output, and when? | “A person reviews anything sent to a customer, published or used in a decision.” |
| Labelling | When do we say AI was used? | “Label content that’s mostly AI-generated. Edited drafts don’t need a label.” |
| AI agents | What may automated agents do on their own? | “Agents need approval before sending money, deleting records or emailing customers.” |
| Questions and incidents | Who do I ask, and what if something goes wrong? | “Ask the named owner. Report mistakes the same day. No blame for reporting.” |
The AI agents row is newer and easy to miss. If you’re using agents that act rather than just write, our guide to human-in-the-loop approval steps covers how to set those limits without slowing everything down.
Rolling it out
- Ask before you write. Find out which tools people already use and for what. You’ll learn what to approve.
- Approve at least one good tool with business terms that cover how your data is handled.
- Write the policy in plain language and keep it short enough that people actually read it.
- Run a 30-minute session with real examples of what’s fine and what isn’t.
- Name an owner and review it yearly, or sooner when you add a new tool.
Rules differ by region. UK and EU businesses have GDPR to consider, US firms face a growing set of state laws, and Gulf businesses have national data laws such as the UAE and Saudi PDPLs. Our article on UK regulators and AI agents covers the British angle in more depth.
If you’d like help choosing tools and setting guardrails, that’s part of our AI strategy consulting. Businesses in Europe can see our UK and Europe page, and you can book a strategy call any time.
This article is general information, not legal advice. Check your obligations with a qualified adviser.
Frequently asked questions
Should we ban ChatGPT and similar tools at work?
How long should an AI use policy be?
Do we have to label AI-generated content?
How often should we update the policy?
Sources
- NCSC blog: The hidden risks of shadow AI (7 Sep 2026)
- Information Commissioner’s Office: Internal AI Use Policy, version 1.3 (PDF)
- Freevacy: NCSC warns against threat of shadow AI (Sep 2026)