Your staff are already using AI. Not the staff you licensed tools for: the marketing assistant pasting client copy into a free chatbot, the estimator asking one to draft a tender answer, the new starter summarising a contract with whatever tool their last employer used. Surveys keep finding a majority of employees using AI at work, most without their employer’s knowledge, and none of it governed by anything.
The risk is not the AI; it is the absence of rules. Company data pasted into free tools may be retained and used beyond your control; confident-sounding wrong answers flow into client work unchecked; and when something goes wrong, “we never told anyone what was allowed” is a poor position for an employer. The fix is one document: an acceptable use policy that says which tools, which data, and who checks what. Here is the template, section by section, UK-framed and sized for SMEs.
The template
Section 1 — Purpose and scope
Two sentences of intent: the business supports productive use of AI tools within rules that protect client data, quality and the company; this policy covers all staff and contractors, on any device, wherever work is done. That last clause matters: the policy governs the *work*, not the office WiFi, or it governs nothing.
Section 2 — Approved tools
A short table: tools the business has approved, the account type required (business accounts, not personal ones, because data handling terms differ), and who owns the licence. Then the standing rules: anything not listed needs approval before use with company data, and the approval route is named. If the business has deployed Microsoft Copilot, it anchors this list, since it processes data inside your tenancy under your controls rather than in someone’s free account.
Keep the list honest and short. A policy that only bans gets ignored; approving good tools is what buys compliance with the limits.
Section 3 — Data rules: what may never go into AI tools
The heart of the policy, and worth stating as bright lines rather than principles:
- Never into unapproved or free-tier tools: client and customer data, personal data of any identifiable person, financial records, credentials or system details, commercially sensitive material (pricing, contracts, plans), and anything under NDA.
- Into approved business-grade tools: as agreed per tool, matching the tool’s data handling terms and your UK GDPR obligations: this is why the approved list carries the burden, and why free-tier accounts are the enemy: many free consumer tiers may use inputs for training, and all of them move your data outside your control.
- The rule of thumb for the grey areas: if you would not email it to an outside consultant without a contract, do not paste it into a chatbot without an approved one.
Personal data deserves its own sentence in the policy: processing it through any AI tool is a UK GDPR processing activity, with all that implies, and goes through the approval route, not individual judgement.
Section 4 — Verification: the human stays responsible
AI output is a draft, not an answer. The policy states it plainly: the member of staff using the tool remains fully responsible for the accuracy, appropriateness and legality of anything they produce with it. Facts, figures, calculations, legal or regulatory statements and anything client-facing get verified before use. “The AI said so” is not a defence the business recognises, and the policy is where that gets established in writing rather than in a post-mortem.
Section 5 — Disclosure
Two audiences, two decisions to pre-make. Internally: staff should be able to say freely that AI helped produce something; secrecy is where unverified output hides. Externally: decide your line by output type (client deliverables, published content, tenders — some of which now ask directly about AI use) and write it down. Whatever your position, it should be a company decision, not fifty individual improvisations.
Section 6 — Security
AI-themed fraud goes both directions, and the policy should acknowledge the incoming kind: voice-cloning and deepfake variants of payment fraud are no longer exotic, and the defence is the same verification-by-second-channel discipline covered in staff security training. Also here: no entering credentials into AI browser extensions or plugins without approval, since that category has produced real incidents.
Section 7 — Breaches and review
What happens when the policy is broken (a proportionate disciplinary reference, aligned with your existing handbook) and, more usefully, a no-blame route for self-reporting: the paste-then-panic moment is recoverable if reported at once, and unrecoverable if hidden. Review the policy every six months: the tool list, the data rules, and whatever the last six months of AI news made newly possible. Date the revisions.
Making it stick
Three implementation notes from watching these succeed and fail. Announce it as enablement, not prohibition: “here is what you *can* use” lands; a ban memo drives usage underground. Pair it with a short training moment at induction and annually. And have leadership follow it visibly, because the fastest way to kill any policy is a director pasting board papers into a free chatbot the week after it ships.
Frequently asked questions
Why do we need an AI policy if we haven’t rolled out AI?
Because your staff have. Ungoverned use of free tools with company data is the default state of most businesses right now; the policy converts that from unmanaged risk into managed practice, whether or not you ever buy a licence.
Should we just ban AI tools instead?
Bans mostly relocate usage to personal phones and home laptops, where you have no visibility at all. An approved-tools policy with bright-line data rules gives you the control a ban only pretends to.
Is pasting client data into ChatGPT a GDPR breach?
Processing personal data through any third-party tool engages UK GDPR, and doing it through an unapproved consumer service with unclear retention is exactly the scenario regulators and clients object to. Business-grade agreements and the approved-tools route exist to make the same productivity lawful.
Does Microsoft Copilot need to be in the policy too?
Yes, as an approved tool with its own note: it operates inside your tenancy under your permissions, which changes the data calculus but not the verification duty. Its prerequisites are their own subject; see our Copilot guide.
Who should own the AI policy in a small business?
Whoever owns your other IT and data policies: typically the operations lead, with your IT provider advising on the tool list and data terms. Six-monthly review dates go in the owner’s calendar, because this subject moves faster than most policies’ annual cycle.
Can we adapt this template ourselves?
Yes, that is what it is for: fill in your tools, name your approval route, align Section 7 with your handbook. If you want the tool-vetting done properly (data handling terms are where the traps live), it is part of what we cover in a health check.
Ship the policy before the incident writes it for you
Every business gets an AI policy eventually; the only question is whether it arrives as a calm document or as the lessons-learned section of an incident report. The template above is free to adapt. If you want the approved-tools list vetted, the Copilot question answered, and the whole thing aligned with your security setup, get in touch and we will fit it into a free IT health check.
