Key takeaways
- Staff are already using AI at work; a short written policy is better than an unwritten assumption.
- Sort tools into tiers by the data they may see, and approve each tool for named tasks.
- Set one checking standard for anything that leaves the firm: a named person reviews it against a source.
- Train before granting access, keep a register, and review the policy at least every six months.
Why a short policy beats none
Most firms already have staff using AI, whether or not the partners have decided anything. The Office for National Statistics found that 55% of UK employees report using AI for work or education, yet only 11% of businesses with ten or more staff say more than half of their workforce has received AI-related training[1]. The gap between use and guidance is where confidentiality slips and unchecked drafts happen.
A policy does not need to be long. Two or three pages that people actually read do more than a twenty-page document nobody opens. Its job is to answer four questions for every member of staff: which tools may I use, for what, with which information, and what checking is required before the output is used.
What holds firms back
Government research found that UK businesses rated ethical concerns (80%), high costs (76%) and unclear or uncertain regulation (72%) as the most significant barriers to adopting AI[2]. A policy will not reduce costs, but it directly addresses the other two: it sets the firm's ethical lines and translates the regulatory position into rules staff can follow.
What the policy must cover
- Scope: who the policy applies to, including consultants, trainees and agency staff, and which devices and accounts.
- Approved tools: where the register is kept and how a new tool is approved.
- Permitted and prohibited uses, by tool tier.
- Information rules: what may and may not be entered into each tier.
- Checking: the standard of human review before any output is used.
- Transparency: what clients and others are told about AI use.
- Training: what is required before access and how often it is refreshed.
- Incidents: how to report errors, near misses and data leaks.
- Ownership and review: who owns the policy and when it is reviewed.
Template clauses to adapt
- This policy applies to everyone who works for or with the firm, on any device or account used for firm work.
- You may use only AI tools listed in the firm's approved tools register, and only for the tasks listed against each tool.
- You must not enter client names, matter details or any confidential information into a tool that is not approved for that information.
- You must not use personal accounts on any AI service for firm work.
- You are responsible for any output you use. Before an AI-assisted document leaves the firm, you must check it against the source material and correct any errors.
- You must not rely on an AI tool for legal research. Every authority must be found and read in an authoritative source.
- AI tools must not be used to make decisions about clients, enquirers or staff without a person reviewing and taking the decision.
- If you are unsure whether a use is permitted, ask the AI lead before proceeding.
- Report any error, near miss or possible data leak involving an AI tool to the AI lead and the COLP on the same day.
- You must complete the firm's AI training before access to any approved tool is granted, and refresh it when asked.
- Where AI materially shapes the service a client receives, the client care information will say so, and you must answer clients honestly if they ask.
- The AI lead reviews this policy and the register at least every six months, and whenever the law, guidance or our tools change.
Sorting tools into tiers
| Tier | Typical tools | Information allowed | Approval | Checking |
|---|---|---|---|---|
| Tier 1: public | Free consumer AI services | None that relates to clients or the firm; public information only | Not approved for firm work | Not applicable |
| Tier 2: contracted general | Business AI services under a firm contract with data protections | Firm information and anonymised client facts | AI lead and data protection lead | Author checks every output |
| Tier 3: contracted with client data | Tools inside your document, case management or office platform, assessed for client data | Client data needed for the task | AI lead, COLP and data protection lead, with a risk assessment | Author checks; supervisor samples |
| Tier 4: client-facing | Chat, call handling or intake tools that speak to the public | Enquirer and client data as designed | Partners, COLP and data protection lead | Scripted, tested and monitored, with escalation to a person |
Tiering stops the policy becoming a list of brand names that goes out of date every month. A new tool is assessed once, placed in a tier and added to the register with the tasks it is approved for. Staff only need to learn the tiers.
Confidentiality and responsibility
Two warnings shape the information rules. The Legal Ombudsman has told consumers that not all free AI tools are secure or keep what users type confidential[3], and the Law Society's generative AI primer, written particularly for small and medium-sized firms, flags misuse or disclosure of confidential information as a key risk[4]. Treat any tool without a firm contract as public.
Responsibility does not move with the work. The SRA is clear that firms remain responsible and accountable for outputs from AI they use, including errors made by a third-party chatbot[5]. The checking clause is therefore the heart of the policy.
Setting the checking standard
Across UK businesses using AI, 84% apply at least some human input or checking to outputs and 67% apply significant checking[2]. For a law firm, significant checking should be the floor for anything that leaves the building. Define it in the policy: a named person reads the whole output, compares it with the source material, verifies any facts, figures, dates and authorities, and takes responsibility for the final version. For internal-only uses such as summarising your own notes, a lighter check may be acceptable, but say so explicitly.
Roles
- AI lead: keeps the register, approves tier 2 tools, runs training and reviews the policy.
- COLP: approves tier 3 and tier 4 tools, receives incident reports and decides on regulatory reporting.
- Data protection lead: checks contracts and data flows, and leads impact assessments where needed.
- Supervisors: know which tools their teams use and sample AI-assisted work.
- Everyone: follows the policy, checks their own outputs and reports problems the same day.
Rolling it out
Find out what is in use
Ask staff anonymously which AI tools they use and for what. Expect surprises.
Draft and consult
Adapt the template clauses, then ask a fee earner, a secretary and a trainee to read it and say what is unclear.
Build the register
Assess the tools you want to keep, assign tiers and list approved tasks.
Train
Run a short session covering the tiers, the information rules, the checking standard and how to report incidents.
Launch and support
Publish the policy, grant access to approved tools and give people a named person to ask.
Review
Check incidents and usage after three months and adjust.
Example: the first register entries
Example (fictional): a nine-person firm's first register has three entries. A contracted business AI service is placed in tier 2 and approved for drafting internal emails, marketing copy and anonymised first drafts. The AI features inside its document platform are placed in tier 3 after a risk assessment and approved for summarising documents already on the matter file. A transcription tool for attendance notes is also tier 3, approved on condition that recordings are deleted once the note is checked. Free consumer tools are listed as tier 1 and not approved for firm work.
Six months later, the review adds one tool, removes a feature the supplier changed without notice, and refreshes training after a near miss in which a draft went to a client before checking. The incident log made that review straightforward.
Keeping it current
Review the policy at least every six months, and immediately when a supplier changes its terms, a new type of tool is proposed, or guidance from the SRA, the Law Society or the courts changes. Keep old versions so you can show what applied at any date. For the regulatory background, read what the SRA expects from firms using AI and AI hallucinations and fake citations. For client-facing wording, see telling clients when you use AI, and to place tools in tiers, use AI tools for legal research and drafting compared.
Frequently asked questions
Does a small law firm need an AI policy?
Yes. Staff are likely using AI already. A short policy setting out approved tools, information rules, checking and incident reporting protects client confidentiality and shows the firm is in control.
What should an AI acceptable use policy include?
Scope, approved tools, permitted and prohibited uses, information rules, the checking standard, transparency with clients, training, incident reporting, ownership and a review date.
Can staff use free AI tools for firm work?
Most firms say no for anything involving client or firm information, because free tools may not keep inputs confidential. Treat any tool without a firm contract as public.
How often should an AI policy be reviewed?
At least every six months, and whenever a supplier changes its terms, a new kind of tool is proposed or regulatory guidance changes.
References
Office for National Statistics (2026). Artificial intelligence in UK businesses: 2023 to 2026 (opens in a new tab) ↑
Department for Science, Innovation and Technology (GOV.UK) (2026). Artificial Intelligence Adoption Research (opens in a new tab) ↑
Legal Ombudsman (2026). AI and complaints: a consumer guide (opens in a new tab) ↑
The Law Society (2026). Generative AI: the essentials (opens in a new tab) ↑
Solicitors Regulation Authority (2023). Risk Outlook report: The use of artificial intelligence in the legal market (opens in a new tab) ↑