Law Society AI Practice Note: Billing, Disclosure, and Checks
The Law Society's AI practice note, dated October 21, 2026, sets expectations on responsibility, verification, client disclosure, and billing.
Quick answer
The Law Society of England and Wales published a practice note on AI in legal practice on October 21, 2026. The note calls itself the Law Society's view of good practice. It says it is not legal advice. It says a solicitor cannot pass professional responsibility to an AI system. It also sets expectations for verification, client disclosure, and billing.
For a U.S. firm, the note is a benchmark. It does not bind your firm. Compare it with your state bar rules before you copy any part of it.
This article is informational and not legal advice. Confirm the rules that apply in your jurisdiction before you change practice.
What happened
The Law Society published "Using artificial intelligence in legal practice" on October 21, 2026. The page is written for all solicitors and all types of legal practice. It covers professional duties, AI governance, confidentiality, privilege, data protection, insurance, and cyber security. The page gives a reading time of one hour and twelve minutes.
The note says there is currently no AI-specific regulator for legal practice. Solicitors use AI under existing duties, including the SRA Standards and Regulations.
The SRA published a warning notice on the misuse of AI in August 2026. The note summarizes its two concerns. The first is false or incorrect information, including citations, in legal work and court submissions. The second is confidential client information entered into AI systems that lack contractual and technical safeguards.
Two points matter most for firm operations.
Responsibility stays with the lawyer. The note says: "Liability cannot be transferred to an AI system as it does not have a separate legal personality." It applies the same rule to AI agents. Their use "does not transfer or diminish professional responsibility."
The note also summarizes a UK Jurisdiction Task Force legal statement. A professional may be negligent for using an unsuitable model, failing to carry out proper due diligence, or failing to validate outputs effectively.
What it means for law-firm workflows
Supervision needs named owners. The note expects responsibility for outcomes to stay "clearly allocated to named individuals within the organisation." For agentic tools, it calls for "clear human oversight and approval checkpoints before actions with legal, regulatory or client impacts are completed." Record the approving lawyer in the matter file.
Verification is a human step. The note says "Any outputs from generative AI need appropriate human verification, oversight and professional judgement." It also says "cannot always be reliably reproduced or independently verified by simply re-running the same query." Running the same query again does not independently verify an answer. Open the source and check it yourself. For the method, see AI Legal Research: A Verification-First Workflow.
Billing has a direct rule. For time-based fees, the note says "only time actually and properly spent may be recorded and billed." It also says not to bill "for the time that a task would have taken had AI not been used." A firm may bill time spent operating the tool, reviewing and verifying its output, and applying professional judgment if the client care letter allows it. The note says any separate AI charge "should be clearly explained and agreed in advance". The firm must not recover that amount twice.
The following example is hypothetical. A lawyer finishes a lease review in forty minutes with AI support. The same review would have taken three hours without it. Under the note, the firm bills the time actually spent, including review. It does not bill three hours. If it adds a technology fee, it agrees that fee with the client in advance.
For a broader view of value and billing, see Law Firm AI ROI: Measure Client Value First.
Client disclosure depends on the use. The note says firms "may encourage clients to disclose any use of AI systems" with them. Firms may also explain AI risks in the engagement terms. Those risks include inaccurate or incomplete material and higher review costs. When a client sends AI-generated material, the solicitor should "review and validate the legal and factual content before relying on the material."
The note also says specific disclosure or consent "may not be required" for internal administrative uses, such as summarizing internal material. It says that using client information that could breach confidentiality, including to train AI, "is likely to require specific client consent."
Vendor terms and features change. The note asks firms to review the provider's terms of service when they adopt a tool. It asks them to check for later changes. It also asks firms to "monitor for AI or AI features being introduced into existing software without prior warning." Firms that run practice management or document software should review feature lists on a set schedule.
Free tools carry more risk. The note says "It is generally riskier to access AI through free, publicly available versions than through enterprise or paid-for licences." It adds that paid tools vary. Their safeguards "will vary by product, contract and specific configuration." Read the contract, not only the product page.
Shadow AI and privilege records need decisions. The note defines shadow AI as use "without the formal approval, visibility or oversight" of IT and security teams. Staff may not see some tools, such as a translation app, as AI at all. The note also asks whether prompts, draft outputs, and audit logs count as lawyers' working papers that attract privilege. It says that question is not yet settled. "Until the position is clarified, you should take a cautious approach."
Practical checks
- Name the lawyer who approves each AI-assisted output, and record that name in the matter file.
- Write a verification step into each workflow. A named lawyer checks the output against the source before it reaches a client or a court.
- Write the billing rule down. Bill time actually spent, including review time. Agree any AI charge in advance, and do not duplicate it.
- Update the engagement letter to explain AI use, the review step, and what you expect from the client.
- Review each provider's terms of service at adoption and after each change. Check where data is stored and processed.
- Ask which tools staff use outside approved systems. Close the gap before a client matter depends on it.
- Decide how your firm treats prompts, drafts, and logs for privilege. Record the decision and the reason.
Limitations and what to monitor
- The note applies in England and Wales. It is guidance, not a rule. This article does not review U.S. ethics rules, state bar opinions, or the ABA's guidance. I could not confirm the current text of those sources in this research window.
- The note leaves some questions open. It says the privilege position is unclear. Artificial Lawyer's commentary asks for a definition of "appropriate" human oversight. I found no definition of that phrase in the note.
- The note says the Law Society will issue further guidance for parts of the profession with distinct needs. Watch for that guidance and for any further SRA material on AI.
- The note runs to more than an hour of reading. This article covers the parts that affect firm operations. Read the full note before you rely on any single point.
Frequently asked questions
Does the Law Society note bind U.S. law firms?
No. It applies to solicitors in England and Wales. It describes itself as the Law Society's view of good practice. U.S. firms should compare it with their own state rules before they adopt any of it.
Can a firm bill for the time AI saves?
Not under this note. Where fees are time-based, the firm bills only time actually and properly spent. It should not bill the time a task would have taken without AI.
Does a firm have to tell clients it uses AI?
The note does not set one rule for every use. It says firms may explain AI risks in engagement terms. It says specific consent may not be needed for internal administrative uses. It says using confidential client information with AI, including for training, is likely to need specific consent.
Can one AI tool check another tool's output?
The note requires human verification and professional judgement for AI outputs. It also says outputs cannot always be independently verified by re-running the same query. A lawyer still checks the source before relying on the output.
Sources and scope
- Law Society of England and Wales, "Using artificial intelligence in legal practice," practice note dated October 21, 2026: lawsociety.org.uk. I read the full page on October 24, 2026.
- Artificial Lawyer, "'AI Is Not A Legal Entity' Says Law Society" (October 21, 2026): artificiallawyer.com. Used only for the reporter's comment on the word "appropriate."
The main source is the Law Society note. The SRA warning notice and the UK Jurisdiction Task Force statement appear in the note as summaries. This article relies on those summaries and does not reproduce either document. The research window ran from October 14 to October 24, 2026. No U.S. authorities were reviewed.
Sources
Keep AI review and time on the matter record
See how Hammer Lex ties time, documents, and review notes to each client matter.
Book a demo