AI Governance and Risk

AI Policy for a Small Law Firm: 8 Clauses You Can Copy

A complete AI policy for a small law firm, free to copy: eight clauses in full wording, each traced to the SRA's August 2026 warning notice, its supervision guidance or the Law Society, plus the test that shows whether your team has read it.

A cut-paper set of scales in a turquoise-to-violet gradient, with a stack of client papers on the left pan and a four-point spark on the right pan, standing for a law firm weighing client confidentiality against its use of AI.

In brief

A small law firm's AI policy needs eight clauses: who it covers, approved tools, client information, citation checks, supervision, clients, mistakes and ownership. Write the client information clause first, because the SRA warns that privilege lost through a public AI tool may be permanently waived. The test of the policy is whether three people at the firm give the same answers.

A small law firm's AI policy needs eight clauses: who it covers, which tools are approved, what client information never goes in, how every citation is checked, who supervises AI-assisted work, what clients are told, what happens after a mistake, and who owns it. In England and Wales, the regulator's August 2026 warning notice makes two of them urgent: client information and citations.

Most small firms already hold the ingredients: a confidentiality policy in the office manual and a supervision routine that mostly happens. What is usually missing is one page that joins the two to AI, in words a paralegal could repeat on the phone. Below is that page, clause by clause, free to copy.

Sources checked November 17, 2026. I am not a lawyer, and this is not legal or regulatory advice. Treat it as a starting draft for whoever is responsible for compliance at your firm.

Does a small law firm have to have an AI policy?

No rule requires a document with that name, but the regulator now expects what the document would contain. The Solicitors Regulation Authority (SRA), which regulates solicitors and law firms in England and Wales, describes its own method as outcomes-focused: "we do not prescribe exactly how those standards should be met in different circumstances." You will not find a rule that says "write an AI policy."

What you will find is a warning notice on the misuse of AI, published on August 17, 2026, which the SRA says is relevant to "all firms and individuals we regulate." It says firms "must also have effective governance structures, systems and controls in place to manage any risks, including those arising from use of AI." And it ends without ceremony: "If you fail to have proper regard to this warning notice, you are at risk of disciplinary action."

The notice was not written in the abstract. In the news release published the same day, the SRA said it had received 42 reports related to the potential misuse of AI between July 2025 and July 2026, with investigations under way on "inaccurate legal citations, supervision and confidentiality."

A written policy is the plainest way for a small firm to show that those systems and controls exist. In a firm of twelve, there is rarely anything else to point to.

What makes a law firm's AI policy different?

Three things set it apart from the policy any other small business needs: privilege, the court and supervision. Everything in the general AI policy template for a UK small business still applies, and if you are still deciding whether your firm needs a policy at all, start there.

Privilege. A shop that pastes a customer's email into a public chatbot has a data protection problem. A law firm that does the same may have given away something that belongs to the client. The warning notice says that using public tools this way "will likely breach client confidentiality and as a result, legal professional privilege may be permanently waived and unable to be recovered." The notice points to an Upper Tribunal decision in an immigration case, UK v Secretary of State for the Home Department [2026] UKUT 81 (IAC), in which the tribunal said: "to put client letters and decision letters from the Home Office into an open source AI tool, such as ChatGPT, is to place this information on the internet in the public domain."

The court. An AI tool can produce a case that does not exist, with a citation that looks right. The SRA expects that "all submissions of named case law authorities are genuine, relevant, have a verifiable citation, and advance the arguments that are being put forward in your documents." It adds: "Reliance on an output of AI would not be a suitable defence in this scenario."

Supervision. The person who used the tool is often not the person whose name is on the file. The notice is direct about who answers for that: "Those who supervise junior or non-authorised colleagues may also be found to have breached regulatory requirements and professional duties if false citations are put before the court without adequate review and/or supervision."

The first of these cannot be undone, so I suggest a firm writes one rule before any other.

The Privilege Line is the rule I suggest a law firm writes first: nothing that identifies a client or a matter goes into an AI tool unless a named person at the firm has confirmed, in writing, that the tool keeps that information inside the firm's control. Every other clause in the policy sits behind that line.

That is my framing, not the regulator's. The regulator's version is longer: "Client information should only be entered into AI systems where appropriate contractual, technical and organisational safeguards are in place to protect confidentiality."

What should a small law firm's AI policy say?

It should say eight things, each short enough to read aloud at a team meeting. Copy the quoted blocks, replace anything in square brackets, and keep the numbering, because people cite clause numbers when they ask questions. The wording follows this publication's US spelling, so change it to your house style.

1. Who this covers

"This policy applies to everyone who does work for [Firm]: partners, consultants, solicitors, trainees, paralegals, secretaries, locums and anyone we outsource work to. It covers any AI tool used for [Firm] work, including free tools in your own browser, tools on your own phone, and AI features inside software we already use."

The free tool on a personal phone is the sentence that earns its place. The Law Society's guidance on generative AI says a solicitor's duties apply "whether AI was used by the solicitor personally or by anyone under that solicitor's supervision."

2. Approved tools

"You may use these AI tools for [Firm] work: [list]. For each one, [name] has read the terms and recorded where our information is stored, how long it is kept, and whether it is used to train the tool. You are encouraged to use approved tools to draft, summarize, plan, check your own writing and prepare for meetings. No other AI tool may be used for client work. To ask for a new one, speak to [name]. The list is reviewed every quarter."

Paying for a tool does not settle the question. The warning notice says "both paid for and free-to-use AI tools may not provide the contractual, and technical safeguards needed to maintain client confidentiality." Reading the terms settles it. Meeting recorders count as AI tools too, and the rule for AI note-takers in client meetings belongs on the same list.

3. Client information

"Nothing that identifies a client, a matter or another party goes into an AI tool unless that tool is on the approved list and [name] has confirmed in writing that it may be used for client information. Nothing about a client ever goes into a free or public AI tool, even with the names removed, if someone could work out who it is about. If you are not sure whether something is confidential or privileged, treat it as both."

This is the Privilege Line as a clause. The Law Society puts the free-tool half bluntly: "If you are using a free, online generative AI service where you have no operational relationship with the vendor other than use, do not put any confidential data into the tool." For approved tools, the SRA lists what [name] is confirming: that client data "remains within a secure environment, is not accessed by unauthorised third parties, is not used to train AI models except where explicitly authorised and appropriate to do so, and is not retained longer than necessary."

4. Citations, quotations and facts

"AI output is a draft. Before any document leaves [Firm], the person whose name is on it checks every case, statute, quotation and factual statement against the source. Checking means opening the law report, the legislation or the client's document. It does not mean asking the tool again. Nothing is filed at court or sent to another party until that check is noted on the file."

5. Supervision

"If you used AI on a piece of work, say so in a file note and tell your supervisor. Supervisors review AI-assisted work before it goes out. For work prepared by trainees, paralegals and anyone who is not authorized, the supervisor opens the authorities personally. A solicitor remains responsible for every piece of work, however it was prepared."

The SRA added sections on AI to its guidance on effective supervision in June 2026. The guidance asks firms to make sure that "outputs produced with the assistance of AI are subject to appropriate human review, scrutiny and professional judgement," and that "an authorised individual retains ultimate responsibility for any legal services delivered with AI assistance."

6. Clients

"Our client care letter says that we use approved AI tools to support our work, that a solicitor reviews everything, and that the client may ask us not to use them. Where a client's own terms restrict AI, those terms come first and you must check before work starts. Any AI tool a client deals with directly is labeled as AI."

This clause is a choice, not a rule I can point to. The Law Society's guidance, last updated in September 2025, says that at that time the SRA had no specific guidance on disclosing AI use for client care, and the warning notice I read does not add a disclosure rule. What the Law Society does say is this: "Clear communication on whether such tools are used prevents misunderstandings as to how information is produced and how decisions are made." For the wider argument, see when to tell customers you use AI.

7. Mistakes

"If AI output that was wrong reaches a client, another party or a court, or if you put something into a tool that this policy says should not go in, tell [name] the same day. Nobody will be penalized for reporting promptly and honestly. [Name] decides who else must be told, which may include the client, our insurer, the court and our regulators."

A policy without this clause teaches people to stay quiet. For the first hour after client information has gone into the wrong tool, use the five-step Paste Plan.

8. Ownership, training and review

"[Name] owns this policy and answers questions about it. Nobody uses an AI tool for [Firm] work until they have been trained on how the tool works and on what our professional duties require, and we keep a record of that training. This policy is reviewed on [date] and whenever our regulator or professional body changes its guidance. Version [1.0], issued [date]."

In most firms the name will be the compliance officer for legal practice. The warning notice reminds firms that this person must take all reasonable steps to make sure the firm complies with its regulatory arrangements, including those on supervision. For a new starter, the first-month AI onboarding plan puts that training into day one, week one and day thirty. The two kinds of training are in the clause on purpose. The Law Society says: "Technical training in the use of such tools is separate from the training in the ethical use of such tools and both are equally important."

Should a small firm ban AI, trust judgment or name its tools?

Name the tools. Here are the three approaches open to a small firm, compared on November 17, 2026. The table is my reading, not a finding from the regulator.

Approach What happens What it costs you
Ban AI for all firm work The rule fits in one sentence Nobody can see whether it is kept: a free tool sits on every phone, and a ban gives staff a reason to say nothing
Leave it to each fee earner's judgment Nothing needs writing Each person draws their own line on client information, and the firm has nothing to show the regulator
Name the tools and write the Privilege Line Staff know what they may use and what never goes in Someone must read the terms of each tool and keep the list current

My verdict: the third. The warning notice says firms "have the freedom to use AI tools and adopt new ways of working, provided they ensure that they meet our standards." The news release quotes the SRA's Aileen Armstrong: "the use of AI does not change the professional standards expected of solicitors and law firms." A ban ignores the first sentence. A named list answers the second.

Who checks the work a paralegal did with AI?

The supervising solicitor does, and "the draft looked fine" has already been tested in court. On May 15, 2026, Legal Futures reported a judgment from a circuit judge sitting in Dudley, who referred two solicitors to the SRA. Appeal documents filed by their firm cited real cases with the wrong citations, or for points they did not establish. The judge said the documents appeared to have been "churned out by a system utilising artificial intelligence." The supervising solicitor told the court that a paralegal had helped prepare them and accepted that he should have checked the research.

One sentence from the judgment is the one for owners: "In my judgment, there were failures here on the part of both solicitors and the management of the firm."

Before you accept "a solicitor reviews it" as the answer, put it through the Loop Test.

"The Loop Test is three questions I ask before accepting 'a human checks it' as an answer: Would that person notice a wrong result without being told to look for it? Would they notice in time to stop it mattering? And if they missed it, could the work be put back? Oversight that survives all three is a safeguard. Oversight that fails any of them is a signature."

Run clause 5 through it. A supervisor notices a wrong citation only by opening the authority. They notice in time only if the check comes before filing. And a document filed at court cannot be quietly put back. That is why clause 4 defines checking as opening the source, and why the SRA's supervision guidance says a supervisor's sample should include "work that the supervisee does not ask them to look at." If your team has never been shown how to check what a tool produces, training staff to check AI output is the place to begin.

What Good Looks Like

Good looks like three people at the firm giving the same three answers. My test for any organization is the Readiness Test. The Readiness Test is three questions an organization must be able to answer consistently, from the top to the front line, before it can call itself AI ready: What is AI for here? What may I do with it? Who decides when it goes wrong?

Ask a partner, a paralegal and whoever opens the post, separately. Clause 2 answers the first question. Clauses 3, 4 and 5 answer the second. Clauses 7 and 8 answer the third. If the three people give three different answers, the firm has a file, not a policy, and the fix is rarely more clauses. It is the person named in clause 8 reading the page aloud, which is what a five-minute team briefing is for.

The organizations I have worked with include Google, Grammarly, Welsh Water and Glasgow City Council, and the test is the same for a firm of ten. Size changes how many people you have to ask. It does not change the questions.

If you run a local law society, a regional network or a professional-practice conference, my business keynotes are about this problem: helping people use AI well and safely in the work they already do.

What if your firm is not regulated by the SRA?

Keep the eight clauses and change the sources. Everything quoted above comes from the regulator and the professional body for solicitors in England and Wales. Firms in Scotland and Northern Ireland answer to their own regulators, so read what those bodies have published before you issue anything.

The same holds for other regulated practices, such as accountants, surveyors and financial advisers. Clause 3 becomes your confidentiality clause, clause 4 becomes your rule for checking figures and references, and clause 8 names whoever answers to your professional body. If you are a solicitor working in-house, the warning notice has a paragraph for you: AI tools built for the wider business "may not necessarily be specifically designed for legal work."

What should you do this week?

A firm of fifteen can do all of this in an afternoon and one phone call.

  1. List what is already in use. Ask everyone, and promise first that nobody is in trouble. That conversation is how shadow AI comes into the open.
  2. Read the terms of the two tools people use most. Write down where information is stored, how long it is kept, and whether it trains the tool.
  3. Fill in the eight clauses. Put one name in the brackets throughout, not a committee.
  4. Ring your insurer or broker. The Law Society's advice to small firms is to "speak to your insurer to determine whether the usage of the tool will be covered under your existing policy." The questions to ask are in does business insurance cover AI mistakes.
  5. Read the policy aloud at the next team meeting. Ten minutes, clause numbers included.
  6. A week later, ask three people the three questions. Change the clause wherever the answers differ.

Where to go next

If the policy is on paper but nobody has shown your fee earners and support staff how to use the approved tools well, that is what my AI training for teams is built for. If the partners have not yet agreed what AI is for at the firm, that is the strategy and governance side of my work with businesses. To see where the firm stands first, the free Business AI Readiness Scorecard takes about five minutes: 12 questions across six areas. Or talk to me about your team.

Privilege, once lost, does not come back. Write the line before someone crosses it.

Sources and further reading

Dan Fitzpatrick is a keynote speaker, Forbes contributor and five-times bestselling author who has trained more than 150,000 people across 30+ countries. About Dan

Key takeaways

  • No SRA rule requires a document called an AI policy, but the August 17, 2026 warning notice expects firms to have governance, systems and controls for the risks of AI, and says ignoring the notice puts a firm at risk of disciplinary action.
  • The SRA received 42 reports related to the potential misuse of AI between July 2025 and July 2026, with investigations into inaccurate citations, supervision and confidentiality.
  • Three things make a law firm's AI policy different from any other small business's: legal professional privilege, the duty to the court, and supervision of junior and non-authorized staff.
  • The Privilege Line: nothing that identifies a client or a matter goes into an AI tool unless a named person has confirmed in writing that the tool keeps that information inside the firm's control.
  • Paying for a tool does not make it safe for client information; the SRA says both paid and free tools may lack the safeguards needed, so someone has to read the terms.
  • Checking a citation means opening the law report or the legislation, not asking the tool again, and the check belongs on the file before anything is filed or sent.
  • A policy works when a partner, a paralegal and the person who opens the post give the same answers to the three Readiness Test questions.

Frequently Asked Questions

Does a law firm legally need an AI policy?

No rule requires a document with that name. The SRA's Code of Conduct for Firms requires effective governance, systems and controls, and its August 2026 warning notice applies that to AI. For a small firm in England and Wales, a short written policy with a named owner is the plainest way to show those controls exist.

What should a small law firm's AI policy include?

Eight clauses: who the policy covers, which tools are approved, what client information may never be entered, how citations and facts are checked, who supervises AI-assisted work, what clients are told, how mistakes are reported, and who owns the policy, trains staff and reviews it.

Can solicitors put client information into ChatGPT?

Not into a free or public tool. The Law Society says not to put any confidential data into a free online AI service, and the SRA warns that doing so will likely breach confidentiality and may permanently waive legal professional privilege. Approved tools need contractual, technical and organizational safeguards first.

Who is responsible if AI invents a case citation?

The solicitor whose name is on the document, and possibly their supervisor. The SRA says reliance on an AI output is not a suitable defense, and that supervisors may breach their duties if false citations reach a court without adequate review. Every authority must be checked against the source.

Do law firms have to tell clients they use AI?

The pages checked in November 2026 set no specific SRA disclosure rule for client care. The Law Society advises clear communication with clients about whether AI tools are used. A sentence in the client care letter, with the client's right to object, is a sensible default, and a client's own terms always come first.

Who should own the AI policy in a small law firm?

One named person, usually the compliance officer for legal practice. The SRA's warning notice reminds firms that this officer must take all reasonable steps to ensure compliance, including on supervision. The owner keeps the approved tool list, answers questions, records training and sets the review date.

Dan Fitzpatrick helps businesses use AI well through keynotes, practical AI training for teams, and AI strategy and governance for leadership teams.

Dan Fitzpatrick

Written by

Dan Fitzpatrick

Dan Fitzpatrick is The AI Educator: a Forbes contributor, international keynote speaker and author who has trained more than 150,000 people across 30+ countries. He works with businesses and leadership teams on AI training, AI strategy and governance, and keynotes.

He also works with schools as an AI keynote speaker for schools, through school AI strategy consulting and with AI training for schools.