Resource: policy template
AI Acceptable Use Policy Template for UK Firms
A clause-by-clause AI acceptable use policy template for UK professional and regulated firms, with the reason for each clause and who owns it.
By Dee Khabra, Founder, The AI Consultancy (London) Ltd.
Published · Reading time approximately 18 minutes.
An AI acceptable use policy tells staff which AI tools they may use, what they may put into them, how output must be checked before anyone relies on it, and who to ask when the answer is unclear. This page is a template for UK professional and regulated firms: fourteen clauses and two schedules a firm can copy and adapt, each with a short note on why it is there, followed by a section on who owns what.
This template is a starting point, not legal advice. Adopting it does not make a firm compliant with any rule, and nothing on this page claims endorsement, alignment, or accreditation from the Solicitors Regulation Authority (SRA), the Financial Conduct Authority (FCA), the Information Commissioner's Office (ICO), ICAEW, or the Law Society. Before the policy is issued, have it reviewed by the people who carry the firm's obligations: the compliance officer for legal practice (COLP), the Head of Compliance, the Data Protection Officer, or an external solicitor.
Last reviewed: 12 November 2026. The sources it relies on are listed at the end, and two of them were still changing on that date.
Does a UK firm need an AI acceptable use policy?
None of the publications discussed below requires a document with that title. Each of them does expect a firm to control how its people use AI, and a written policy is a direct way to do that.
The Law Society's practice note "Using artificial intelligence in legal practice" (21 October 2026) puts the position plainly for solicitors: "Your organisation is not required to have a standalone internal AI use policy." It then adds that an organisation "should communicate what is, and is not, acceptable AI use", and that this is needed even where the organisation has not formally adopted AI, because staff may be using free public tools on their own. The note calls that shadow AI.
The SRA's warning notice "Misuse of AI" (17 August 2026) applies to every firm and individual the SRA regulates. It says firms must have effective governance structures, systems and controls in place to manage risks, including those arising from the use of AI, and it cites paragraph 2.1 of the Code of Conduct for Firms. The notice says a firm that fails to have proper regard to it is at risk of disciplinary action.
The FCA's "AI in financial services: good practice and areas for improvement" (19 October 2026) covers regulated firms that develop, deploy, procure or use AI, including tools "used by staff to support decision making". The FCA says its examples are illustrative and "do not create new requirements". They do show what the regulator has seen work.
ICAEW's Code of Ethics asks members who use the output of technology to exercise professional judgement about whether relying on it is reasonable. Its helpsheet "The impact of technology" lists the factors, and one of them is the controls relating to the use of the technology, including procedures for authorising user access.
A policy does not discharge these duties. It gives staff one set of rules to follow, and it gives the firm something to train against and to show a client, an insurer, or a regulator who asks how AI is controlled.
How should a firm use this template?
Copy the clause wording, replace the text in square brackets, and delete what does not apply. Keep the finished policy to two or three pages, so that staff can read it in one sitting.
A policy is only useful if three things are true:
- The list of approved tools is real and current. If the firm has approved nothing, staff may use whatever is free.
- The rule on what may be entered into a tool is specific enough to apply at a desk without asking anyone.
- A named person owns the policy and answers questions about it within a working day.
The policy covers every kind of AI the firm uses. The Law Society's note says an AI policy "should not be limited to generative AI only", and it notes that where AI is embedded in existing software a firm may not have clear visibility of where it is operating.
What should the policy say? Fourteen clauses
Each clause below gives wording to copy, then the reason for it.
1. Purpose and scope
This policy sets out how [firm name] uses artificial intelligence (AI) in its work. It applies to all partners, directors, employees, contractors, and temporary staff, and to any AI tool used for the firm's work, whether the firm provides the tool or not. It applies to AI features inside other software as well as to standalone AI tools.
Why it is there: a policy that covers only the tools the firm bought leaves a gap. The scope has to reach personal accounts, free tools, and AI features switched on inside existing products.
2. Definitions
"AI tool" means any software that generates text, images, audio, code, summaries, scores, or recommendations that were not explicitly programmed in advance. "Approved tool" means an AI tool listed in Schedule 1. "Confidential information" means any information about a client, a matter, the firm, or a person that is not publicly available. "Personal data" has the meaning it has in UK data protection law.
Why it is there: staff cannot follow a rule about confidential information if they have to guess what counts. The definition used here follows ICAEW's Code of Ethics, which defines confidential information as any information, data or other material that is not publicly available. A law firm should add legally privileged material to the definition by name.
3. Approved tools
You may use only the approved tools in Schedule 1 for the firm's work, and only for the uses listed against each tool. You must use them through the account the firm gives you. [Policy owner] keeps Schedule 1 up to date and reviews it every [three] months.
Why it is there: the SRA's notice says "both paid for and free-to-use AI tools may not provide the contractual, and technical safeguards needed to maintain client confidentiality". The firm, not the individual, has to check those safeguards, so the firm has to choose the tools. Schedule 1 is the firm's AI register in its simplest form. The Law Society's note suggests a register records the approved systems, the approved uses, the owner of each system, and the dates of approval and review.
4. Tools and accounts that are not approved
You must not use an AI tool that is not in Schedule 1 for the firm's work. You must not use a personal account, a free version, or a trial of any AI tool for the firm's work, even where the firm has approved a paid version of the same product. If you think a tool should be approved, ask [policy owner] under clause 11.
Why it is there: the SRA's notice says that, depending on the provider's terms, settings and technical architecture, information entered into an AI system may be stored, retained or used to improve the tool. Those terms can differ between the firm's account and a personal or free one. Naming personal accounts closes a common route for shadow AI. The last sentence gives staff a way to ask, so that a ban is not the only answer they get.
5. Information you must not enter
You must not enter any of the following into an AI tool unless Schedule 1 says that tool is approved for it: confidential information about a client or a matter; personal data; special category data, such as health information; [legally privileged material]; [inside information or information the firm holds under a non-disclosure agreement]. If you are not sure whether information falls into one of these groups, treat it as if it does and ask [policy owner].
Why it is there: this is the clause the SRA's notice bears on most directly. The notice says "client information should only be entered into AI systems where appropriate contractual, technical and organisational safeguards are in place to protect confidentiality", and that firms should satisfy themselves that client data stays in a secure environment, is not accessed by unauthorised third parties, is not used to train AI models except where explicitly authorised, and is not kept longer than necessary. For law firms the stakes include privilege. The notice records the Upper Tribunal's observation that putting client letters into an open tool places the information in the public domain, and says privilege may be permanently waived as a result.
6. Checking output before you use it
AI output can be wrong, out of date, or invented, and it can look convincing when it is. Before you rely on AI output or send it to anyone, you must check it against a source you trust. You must check every case, statute, citation, quotation, figure, and calculation against the original. You must not present AI output as checked when it has not been.
Why it is there: the SRA's first stated concern is court and other documents containing false or incorrect information, including citations, produced by AI. The same risk applies to a figure in a set of management accounts or a statement in a suitability report. The clause lists what to check because "review the output" is too vague to supervise.
7. Responsibility and supervision
You are responsible for work you produce with an AI tool in the same way as for any other work. Using an AI tool is not a defence if the work is wrong. If you supervise others, you are responsible for checking that they follow this policy, and your review of their work must cover the parts an AI tool produced.
Why it is there: the SRA's notice says "AI has no separate legal personality" and that those who use it remain accountable for their work regardless of how it was prepared. It also says supervisors may breach their own duties if false material goes out without adequate review. In financial services, the FCA's publication describes human review of AI outputs becoming a "box-ticking" exercise as an area for improvement.
8. Decisions about people
You must not use an AI tool to make a decision that has a legal or similarly significant effect on a person, such as a decision on recruitment, promotion, dismissal, credit, pricing, or eligibility for a service, unless [Data Protection Officer] has approved that use in writing. Where an AI tool informs a decision of this kind, a person with the authority to change the outcome must consider the case and make the decision.
Why it is there: the UK GDPR's automated decision-making rules (Articles 22A to 22D, which replaced Article 22 under the Data (Use and Access) Act 2025) apply where a significant decision is taken with no meaningful human involvement. Such decisions need safeguards: the person must be given information about the decision and be able to make representations, obtain human intervention, and contest it.
The ICO's guidance says you should carry out a data protection impact assessment for any solely automated decision with a significant effect. That guidance was still a draft under consultation in November 2026, so check the current version. Routing these uses to the Data Protection Officer keeps the judgement with the person equipped to make it.
9. Telling clients and others
You must not mislead a client, a court, a regulator, or anyone else about whether or how AI was used in a piece of work. If a client asks how AI is used on their work, answer accurately or refer the question to [named role]. Where a client's terms or instructions restrict the use of AI, those terms apply to that client's work and override Schedule 1.
Why it is there: the Law Society's note says providing appropriate information to clients includes not misrepresenting the use of AI. It also suggests a firm may set out its use of AI in its terms of engagement. The final sentence covers client terms: a client's outside counsel guidelines or master services agreement may be stricter than the firm's own policy.
10. Records
Where you use an approved tool on [client work or a decision listed in Schedule 2], record on the file which tool you used, what you used it for, and who checked the output.
Why it is there: a firm that is asked later how a piece of work was produced needs to be able to answer. The FCA's publication describes a small firm that emphasised the need to record whether staff needed to review AI outputs, whether the review happened, and whether they approved the output, overrode it, or escalated it. The FCA notes that this created an audit trail of how oversight worked in practice. Keep the record short enough that people will keep it.
11. New tools and new uses
Before the firm adopts a new AI tool, or uses an approved tool for a new purpose, [policy owner] must approve it. The request must say what the tool will be used for, what information will go into it, and who the supplier is. [Policy owner] will consult [Data Protection Officer] and [IT or security lead], and will decide whether a data protection impact assessment is needed.
Why it is there: the FCA's publication lists retrospective governance, where a firm deploys AI and then tries to fit it into existing controls, as an area for improvement. A short approval step before a tool is used prevents it. Our article on AI procurement due diligence sets out the questions to put to a supplier.
12. Training
You must complete the firm's AI training before you are given access to an approved tool, and refresher training every [twelve] months. Supervisors must complete additional training on reviewing work that an AI tool has contributed to.
Why it is there: the SRA's Codes require firms to ensure their people are competent and keep their knowledge up to date. ICAEW's Code says maintaining professional competence requires "a continuing awareness and understanding of technical, professional, business and technology-related developments relevant to the professional activities undertaken". The FCA's publication describes generic or awareness-only training in oversight functions as an area for improvement, which is why the clause separates supervisors from general users.
13. Reporting concerns and incidents
If you think information has been entered into an AI tool in breach of this policy, or that incorrect AI output has been sent outside the firm, tell [named role] straight away. Reporting promptly will be treated as a mitigating factor. [Named role] will decide whether the incident must be reported to a client, an insurer, or a regulator.
Why it is there: the ICO's guide to personal data breaches says a notifiable breach must be reported to the ICO without undue delay, and not later than 72 hours after the organisation becomes aware of it. The firm therefore needs to hear about an incident on the day. The clause says prompt reporting counts in a person's favour so that staff have less reason to stay quiet.
14. Ownership, review, and breach
[Policy owner] owns this policy. It was approved by [the board or the partners] on [date] and will be reviewed every [six] months and whenever the law, the firm's regulators' guidance, or the approved tools change. Breach of this policy may be dealt with under the firm's disciplinary procedure.
Why it is there: guidance in this area is moving. The Law Society's note makes the point that a standalone policy can be updated quickly without reopening the firm's whole technology framework, and a review date is what makes that happen.
Schedule 1: approved tools
For each approved tool, record: the tool and the plan the firm pays for; the uses it is approved for; the kinds of information that may be entered, and any that may not; the internal owner; the date of approval; and the date of the next review.
Why it is there: clauses 3, 4, and 5 all point here, so this is the page staff will open most. A firm that keeps an AI register can make Schedule 1 an extract from it.
Schedule 2: uses that need a record
A record under clause 10 is required for: any use on client work; any use that contributes to a document for a court, a tribunal, or a regulator; any use that informs a decision about a person; [any use that affects customers directly].
Why it is there: the list limits the record-keeping duty to the uses where the firm would later need to show what happened.
Which clauses change with the firm's regulator?
The fourteen clauses apply across sectors. Four adjustments depend on who regulates the firm.
A law firm should name legally privileged material in clauses 2 and 5, name the COLP in clause 13, and add a line to clause 6 that nothing is filed at court or sent to another party until a solicitor has verified every authority in it. The AI for UK law firms article covers the SRA's notice in more detail.
An FCA-regulated firm should make the policy owner in clause 14 answerable to the senior manager responsible for the business area, and list in Schedule 2 the uses that affect customers. The FCA's publication says it does not expect senior managers to be technical specialists, but that they should take reasonable steps to understand the AI risks in their area. See AI in UK financial services.
An accountancy practice should tie clause 6 to its existing file review, and should add to clause 9 that the firm will explain to a client the implications of any limitations in the technology used on their work. ICAEW's helpsheet says the Code obliges members to explain the implications of limitations inherent in their professional activities, including limitations related to technology. See AI for UK finance teams.
Any firm that uses AI in hiring or in decisions about customers should expand clause 8 with the specific uses the Data Protection Officer has approved. See AI in UK hiring.
Who owns what?
A policy fails when everyone assumes someone else is running it. These are the roles to name before it is issued.
- The board or the partners approve the policy and set how much risk the firm will accept. The Law Society's note says accountability for an organisation's approach to AI should sit with its leadership.
- The policy owner is one named senior person. They keep Schedule 1 current, decide requests under clause 11, and answer questions. In a law firm it could be the COLP or a partner working with the COLP. In an FCA-regulated firm it should be someone who reports to the responsible senior manager.
- The Data Protection Officer or data protection lead decides on uses that involve personal data, approves anything under clause 8, and decides whether a data protection impact assessment is needed.
- The IT or security lead checks each tool's settings, access controls, and retention, and removes access when people leave.
- Supervisors review work that AI contributed to and are the first point of contact for their teams.
- Every member of staff follows the policy, checks their own output, and reports incidents.
- HR or the learning lead keeps the training records that clause 12 depends on.
In a firm of thirty people, one person may hold three of these roles. That is fine as long as each role has a name against it.
What does the template not do?
It does not replace the firm's own assessment of its tools. Schedule 1 is only as good as the due diligence behind each entry.
It does not cover building or training AI models, which raises further questions about data, testing, and intellectual property.
It sets rules for staff. It does not set out how the firm decides which uses of AI to allow in the first place, who signs them off, or how they are monitored. That is the job of a governance framework, and our AI governance framework for UK regulated firms covers it.
It is not a substitute for training. A policy tells people the rule. Training shows them how to apply it to the work on their desk. The AI governance briefing for regulated firms works through these questions with a firm's compliance, legal, and data protection leads, and the free AI Readiness Assessment shows where a firm stands today.
Sources
Each source was read at the publisher's own site on 12 November 2026.
- Solicitors Regulation Authority, Misuse of AI (warning notice), 17 August 2026.
- Financial Conduct Authority, AI in financial services: good practice and areas for improvement, 19 October 2026.
- Information Commissioner's Office, Automated decision-making, including profiling, draft guidance updated 31 March 2026 and under consultation.
- ICAEW, The impact of technology (Code of Ethics helpsheet), published 27 January 2025. A new edition of the Code came into force on 1 July 2026.
- Information Commissioner's Office, Personal data breaches: a guide.
- The Law Society, Using artificial intelligence in legal practice (practice note), 21 October 2026. The Law Society says it is updating the note.