Tekmait

Free AI policy template for UK small firms

No UK law makes a small firm write an AI policy. But data protection law and professional codes from the SRA, ICAEW and the tax bodies expect you to control which AI tools staff use and what client data goes into them. This free template turns those duties into rules your team can follow.

Free. No email needed.

Checked on 11 November 2026. Written by Gary Dhanda.

This page is general information, not legal advice. Check your own obligations with your regulator or a qualified adviser.

Is an AI policy a legal requirement in the UK?

We have found no UK statute that requires a small firm to write an AI policy. The Law Society says the government chose a light-touch approach in its February 2024 white paper response and left existing regulators to apply existing rules. UK GDPR Article 24(2) does expect appropriate data protection policies where that is proportionate. Professional bodies expect firms to set rules for AI use.

What should an AI usage policy include?

Ten parts: purpose and scope, approved tools and plans, data staff must not enter, memory and chat history settings, checking outputs, telling clients, connected tools and agents, incidents, training and an owner with a review date. The template below has all ten.

AI use policy for [Firm name]. Owner: [Name]. Approved: [Date]. Next review: [Date].

1. Purpose and scope. This policy sets how everyone at [Firm name] uses AI tools such as ChatGPT, Claude, Copilot and Gemini for work. It covers staff, contractors and anyone using firm data.

2. Approved tools and plans. Staff may use only these tools on these plans: [Tool, plan, who holds the account]. Work must not be done on personal accounts or free plans unless the owner has approved it in writing.

3. Data we do not enter. We do not enter client names, identifiers, financial details, health information, passwords or privileged documents into any tool that is not on the approved list. Where a tool is approved, we enter the least data needed to do the task.

4. Memory and chat history. On approved tools we set memory and chat history as follows: [Tool, memory on or off, history on or off, who checked, date]. Staff do not ask a tool to remember client details. We review what each tool has saved every [three] months and delete anything that should not be there. When someone leaves, the owner reviews their AI accounts and saved memories the same day.

5. Checking outputs. A person checks every AI output before it is used or sent. We check facts, figures, names, citations and tone. Nobody relies on an AI output alone for advice, a filing or a decision about a person.

6. Telling clients. We tell clients when AI has been used on their work where our professional rules or engagement terms require it: [Wording].

7. Connected tools and agents. We do not connect an AI tool to email, files or other systems unless the owner has approved it. Anything it reads can contain instructions, so we treat its output as untrusted and limit what it is allowed to do.

The NCSC says current large language models do not enforce a security boundary between instructions and data in a prompt and that prompt injection may never be fully mitigated.

8. Incidents. If client or personal data is entered into the wrong tool, tell [Name] the same day. We record what happened, delete what we can, check the tool's retention settings and decide whether the matter must be reported.

9. Training. Everyone is briefed on this policy when they join and when it changes. Date of last briefing: [Date].

10. Owner and review. [Name] owns this policy and reviews it at least every six months and whenever we change a tool or plan.

Can staff put client data into ChatGPT, Claude or Copilot?

It depends on the plan and the data. ICAEW guidance says not to load confidential information into public generative AI tools. The SRA says firms need clear rules for staff use of ChatGPT-type tools even if the firm has not adopted AI. A workable rule is approved tools and plans only and no client names, identifiers or financial details anywhere else.

What each vendor says about training on your data (checked 11 November 2026)
VendorPersonal plansBusiness plans
OpenAIOpenAI says it may use content you provide to ChatGPT, including memories, to improve its models. Check the data controls setting on each personal account.Business data is excluded from training by default (OpenAI, ChatGPT Business and Enterprise pages).
AnthropicFree, Pro and Max users choose. If training is allowed, data is kept for five years, otherwise 30 days (Anthropic, consumer terms update, 28 August 2025).The change does not apply to Claude for Work (Team and Enterprise) or the API.
GoogleNot covered here.Google says Workspace content is not human reviewed or used for generative AI model training outside your domain without permission.
MicrosoftNot covered here.Microsoft says prompts, responses and data accessed through Microsoft Graph are not used to train foundation models behind Copilot.

Vendor pages change. Check the plan you actually use.

What does the ICO say about using generative AI with personal data?

UK data protection law applies when personal data goes into an AI tool. The ICO publishes guidance on AI and data protection and an AI risk toolkit and its AI and biometrics strategy says it will develop a statutory code of practice on AI and automated decision-making. Read the ICO's current pages and their dates before you rely on any summary, including this one.

ICO

Does the Data (Use and Access) Act 2025 change anything for small firms?

Not much for everyday drafting and summarising. The Act replaced the automated decision-making rules in Article 22 with new Articles 22A to 22D. They matter if AI output decides something significant about a person, such as recruitment screening or eligibility, with no meaningful human involvement. If you do that, take advice first.

What do the Law Society, SRA and ICAEW expect from an AI policy?

  • Law Society: its generative AI guide asks firms to define the purpose, review how the vendor handles data, check whether a tool is a closed system or also trains models for third parties, tell clients, record inputs and outputs and check results.
  • SRA: the firm stays responsible for outputs and needs clear rules and staff training.
  • ICAEW: in November 2026 it told firms to make sure internal AI policies, procedures and training reflect the Code of Ethics, including confidentiality (R114.1).
  • Tax bodies: seven bodies, including AAT, ACCA, CIOT and ICAEW, published AI guidance for tax work on 19 January 2026.

The government's AI Management Essentials self-assessment tool did not launch as a final tool. DSIT's December 2025 consultation response says it will develop guidance aimed at SMEs instead.

How do we stop AI tools remembering client information?

Use business plans and set memory at admin level where the vendor allows it. ChatGPT Enterprise account owners can turn memory off for the organisation. Claude organisation owners can turn memory off for their organisation. In Microsoft 365 Copilot, an admin can turn off enhanced personalization, and users then cannot turn on their own custom instructions, saved memories or chat history. Microsoft also says deleting a chat does not delete memories created from it, so check saved memories as well as chats. Our AI Memory Audit does this for you.

AI Memory Audit

How often should we review our AI policy?

At least every six months (our suggestion) and whenever you change a tool or plan, because vendors change defaults. Anthropic, for example, made memory on by default for most Team and Enterprise organisations created on or after 31 October 2026.

Related

Sources