This is a plain-English summary of published guidance, not legal advice.
What should an AI policy include?
An effective policy covers eight areas, each mapped to existing SRA duties.
- Approved tools. Which tools people may use, for what, and who approves new ones.
- Client data. Which data is allowed in approved tools, and what is strictly excluded.
- Human review. Who checks AI output, what they check and how it's recorded.
- Supervision. Who is accountable for AI-assisted work, and the COLP's role.
- Telling clients. What your client care letter says, and what happens if a client says no.
- Training. What people learn before they get access, and how you record it.
- Incidents. Who to tell, and when the client, the ICO or the SRA needs to know.
- Review. Who owns the policy and how often it changes.
The SRA AI guidance page explains where each duty comes from.
Questions to ask when drafting an AI policy
Answer these with your COLP before you edit the template. The answers fill in most of the brackets.
- Which AI tools are people already using, including on their phones?
- Which tools will we approve, and which ones are cleared for client data?
- What must never go into an AI tool at this firm?
- Who checks AI-assisted work, and how do we record that on the matter?
- What do we tell clients, and have any clients already told us not to use AI?
- Who approves a new tool, and what do they check first?
- What happens when someone makes a mistake, and who decides whether to report it?
- Do our insurers want to know how we use AI?
- Who owns the policy, and when is the next review?
If several of these have no answer yet, score the firm against our AI readiness checklist first. It shows the gaps in data, systems, people, policy and UK regulation that the policy needs to cover.
The AI acceptable use policy template
Replace the text in square brackets and delete anything that doesn't fit your firm. Copy it into Word, or download it as a text file.
- Owner: [COLP name]
- Approved by: [Managing partner name]
- Effective from: [date]
- Next review: [date, no more than six months away]
1. Purpose and scope
- 1.1This policy sets out how everyone at [Firm name] may use artificial intelligence (AI) tools for firm work, including generative AI such as chat assistants, drafting tools and call transcription.
- 1.2It applies to partners, solicitors, trainees, paralegals, support staff, consultants and contractors, on firm devices and on any personal device used for firm work.
- 1.3It applies alongside our information security, data protection, supervision and complaints policies and the SRA Standards and Regulations. Where they conflict, the stricter rule applies.
- 1.4We use AI to save time on routine work, while a person at the firm remains responsible for all work we produce.
2. Approved tools
- 2.1Only use AI tools on the approved list for firm work. The list is kept at [location] by [role].
- 2.2For each tool, the list records what it may be used for, what data may go into it, who may use it and where its data is stored.
- 2.3[COLP or IT lead] only approves a tool once they have confirmed that the provider does not train its models on our data, that data is held securely and deleted on our instruction, that we have a data processing agreement, and where data is stored and processed.
- 2.4Do not use personal or free AI accounts (for example a personal ChatGPT, Claude or Gemini login) for client or firm work.
- 2.5To ask for a new tool, email [role] with what you want it for. Don't try it with client or firm data while you wait.
3. Client and firm data
- 3.1Treat anything you type, paste or upload into an AI tool as shared with the tool's provider.
- 3.2Client information, including names, matter details, documents and correspondence, may only go into approved tools that the list marks as cleared for client data.
- 3.3Never put the following into any AI tool unless the approved list expressly allows it: special category personal data; material covered by a court order, undertaking or confidentiality agreement; anything about a suspicious activity report; [add your own].
- 3.4Where the task can be done without identifying details, remove names and identifying details first.
- 3.5Privileged material only goes into approved tools that run inside our own systems [for example, our Microsoft 365 tenant]. If you are not sure a tool is safe for privileged material, don't use it.
- 3.6Some clients restrict or forbid AI on their matters. Check the engagement terms and the matter file before you start.
4. Human review
- 4.1AI output is a first draft. A suitably qualified person must review it before it is relied on, sent to a client or another party, or filed with a court or tribunal.
- 4.2The reviewer checks facts, dates, figures and reasoning against the source documents, and checks every case, statute and quotation against a reliable primary source. Never ask an AI tool to check its own work.
- 4.3Do not put any document before a court or tribunal that cites an authority you have not personally checked.
- 4.4Record on the matter file when AI was used for substantive work, which tool was used and who reviewed the output.
- 4.5AI does not make decisions about clients, such as whether to take on a matter, whether a conflict exists or the merits of a claim. It can prepare information for the person who decides.
5. Supervision and responsibility
- 5.1The fee earner with conduct of a matter is responsible for AI-assisted work on it, as with any other work.
- 5.2Supervisors must know which AI tools their team uses, and check AI-assisted work by trainees, paralegals and support staff in the same way as other work.
- 5.3[COLP name] is responsible for this policy, the approved list and the firm's compliance with both. [Name or role] is responsible for data protection.
- 5.4Anyone can raise a concern about AI use with their supervisor or the COLP, and will not be blamed for doing so.
6. Telling clients
- 6.1Our [client care letter or terms of business] tells clients that we may use AI tools on their matter, that a qualified person reviews the work, and how we protect their data.
- 6.2Where a client deals with an AI tool directly, such as an enquiry assistant, we make that clear and give them a way to reach a person.
- 6.3If a client asks how AI was used on their matter, we answer honestly.
- 6.4If a client tells us not to use AI on their matter, we follow that instruction.
- 6.5We describe AI-assisted work fairly on bills. [Set out how AI-assisted work is reflected in your fees.]
7. Training
- 7.1Everyone completes [the firm's AI training] and confirms they have read this policy before they get access to an approved tool.
- 7.2Training covers what each approved tool is for, what data may go into it, how to check its output, how AI gets things wrong (including invented cases and citations) and how to report a problem.
- 7.3[Role] keeps a record of who has been trained and when. Training is refreshed [every year] and whenever we add a tool.
8. Incidents
- 8.1Tell [COLP name] straight away if client or confidential information has gone into an unapproved tool, if AI output with an error has reached a client, another party or a court, or if a tool has behaved unexpectedly with firm data.
- 8.2The COLP records the incident, assesses the harm and decides what happens next, including whether to tell the client, notify our insurers, report a personal data breach to the ICO (within 72 hours where required) and report to the SRA where our reporting duties apply.
- 8.3Honest mistakes reported promptly are treated as a chance to learn. Hiding an incident is a disciplinary matter.
9. Breaches of this policy
- 9.1Breaching this policy may lead to disciplinary action. It may also breach your obligations under the SRA Standards and Regulations.
10. Review
- 10.1[COLP name] reviews this policy and the approved list at least every [six] months, and sooner when we add a tool, after an incident, or when the SRA, the ICO or the courts publish something new on AI.
- 10.2Changes are recorded in the version history below and shared with everyone at the firm.
- 10.3Version history: [version, date, what changed, approved by].
- 10.4Acknowledgement: I have read and understood this policy. [Name, signature, date].
How to roll it out
- Get the managing partner and COLP to agree the approved list.
- Turn off or block the tools that aren't on it, where your IT provider can.
- Add the AI wording to your client care letter.
- Run a short session for everyone, then collect signed acknowledgements.
- Put the review date in the COLP's diary.
For more detail on what the regulator expects, read the SRA's warning notice on misuse of AI and its compliance tips on AI and technology.
Want AI set up to match the policy?
Staff stick to a policy when the approved tools are convenient to use. We configure AI inside your practice management system and Microsoft 365 so client data stays in your tenant and fee earners review all client outputs. We also draft a tailored AI policy, with prompts and client wording, for your COLP to adapt and own.
