UK: what the SRA and judicial guidance say about AI use
By the ChamberOne team 5 minute read
Solicitors in England and Wales have no single rulebook for AI. What exists is a set of published positions from the regulator and the judiciary, each aimed at a different audience. This article summarises two of them, the Solicitors Regulation Authority's Risk Outlook report on AI and the judiciary's guidance for judicial office holders, and draws out what they suggest for a firm. It is a summary of what has been published, not legal advice, and both documents should be read in full.
Key points
- The SRA's Risk Outlook report says firms remain accountable for AI outputs and cannot delegate that to a provider.
- It flags accuracy, confidentiality, bias and scale as risks, and cites AI-drafted arguments that included non-existent cases.
- The judiciary's guidance is addressed to judicial office holders, not solicitors, but shows what courts expect.
- Practical steps include a written policy, a named verifier for court and client documents and staff training.
- Check the SRA and judiciary websites for current versions before relying on any summary.
Two documents, two audiences
The SRA is the regulator of solicitors in England and Wales. In November 2023 it published a Risk Outlook report called The use of artificial intelligence in the legal market. The report discusses opportunities and risks for firms and sets out matters to consider when adopting AI.
The judiciary of England and Wales has separately published guidance on AI for judicial office holders. It is addressed to judges and those who support them, not to solicitors. The version current at the time of writing is dated October 2025 and replaces an April 2025 version.
What the SRA report says about accountability
The central message is that firms remain responsible for the outputs of AI. The report states that firms cannot delegate accountability to an IT team or an external provider, and that if a third-party chatbot gives initial advice, the firm is still answerable for errors in it. It also says firms must ensure clients are told how AI is involved in their matters.
For a firm, that means supervision does not stop where a tool begins. Whoever signs the work owns the result.
This applies equally to a trainee who used a tool for a first draft and to a partner who relied on a colleague's AI-assisted note. The question to ask of any piece of work is who has checked it and how, not which tool produced it.
What it says about risks
The report identifies several risks. These include accuracy, because generative models predict plausible text without understanding what is true; it cites AI-drafted legal arguments that included non-existent cases, and notes that people may place more trust in computers than in humans. It also covers bias, confidentiality, scale, crime such as deepfakes, and differences in rules between jurisdictions.
On confidentiality, the examples include staff entering client details into public AI tools and confidential data being exposed when shared with a provider.
The report also touches on the speed at which a tool can produce material. Because output can be created quickly and in volume, supervision becomes harder, and a single error can be repeated across many documents before anyone notices.
What it suggests firms consider
The report sets out practical matters for firms. Treat the list as a prompt for your own policy.
- Select tools that fit the firm's needs, understand where responsibility for errors lies and test systems before use.
- Set clear rules on acceptable use, including casual use of public tools, and train staff to question AI effectively.
- Check outputs for accuracy, and do not rely on the AI system to verify itself.
- Understand how systems reach their results and be able to explain AI use to clients.
- Limit the data shared and take care over where cloud-based data is held.
- Monitor outputs for bias and for changes over time.
- Discuss AI use with insurers.
What the judicial guidance says
The judiciary's guidance applies to judicial office holders and their clerks, judicial assistants, legal advisers and other support staff. Its themes are familiar. Judicial office holders should not enter private information into public AI tools. AI tools can produce information that is inaccurate, incomplete, misleading or out of date, and the guidance expands on the risks of bias and hallucination. Judicial office holders remain personally responsible for material produced in their name.
The guidance is not a rulebook for solicitors. Its value for practitioners is as a statement of what courts will expect to see: care with confidentiality, and verification of anything an AI tool produces.
Read as a pair, the two documents describe the same habits from different seats: the regulator asks firms to take responsibility and show their checks, and the judiciary asks its own members to protect confidentiality and to remain personally responsible for what is issued in their names.
- Addressed to firms.
- Firms remain responsible and cannot delegate accountability.
- Clients should be told how AI is involved.
- Flags accuracy, confidentiality, bias and scale.
- Addressed to judicial office holders and support staff.
- No private information in public AI tools.
- Output can be inaccurate, incomplete or out of date.
- Judges remain personally responsible for what they issue.
What this implies for supervision
Taken together, the documents point in one direction. Responsibility stays with the people and the firm. That suggests a few concrete steps, which are our reading and not a statement of what any rule requires.
- Know which AI tools are in use in the firm, including informal ones.
- Have a written policy covering approved tools, prohibited inputs and checking.
- Make a named lawyer responsible for verifying anything that goes to a court or a client.
- Train everyone who touches client information, not only fee earners.
- Keep a record of how AI was used on important matters.
- Know which tools are in use, including informal ones.
- Have a written policy on tools, inputs and checking.
- Name a lawyer to verify court and client documents.
- Train everyone who touches client information.
- Delegate accountability to an IT team or provider.
- Rely on the AI system to verify itself.
What this implies for disclosure
The SRA report says firms must ensure clients are told how AI is involved. How and when to tell a client, and whether to tell a court, is something to decide with reference to the SRA's current standards and guidance, and any directions from the court in which you appear. This article does not say what any rule requires.
A reasonable practice is to include a short, plain statement in engagement terms and to be ready to explain use of tools on request.
If the firm uses AI only for internal tasks that a lawyer then reviews, the statement can be brief. If client information is processed by a third-party tool, the statement should say so in terms the client can understand, and the firm should be able to answer follow-up questions about where the information goes.
Keep reading, because this moves
Both documents have been updated or supplemented, and others may exist from the Law Society, the Bar Council and the courts. Check the SRA's website and the judiciary's website for the current positions before you rely on this summary. Where you are unsure whether a use is appropriate, speak to your firm's compliance officer or seek independent advice.
Sources
This article is general information about the law, legal practice and technology, current when it was published. It is not legal advice. Laws and guidance change: check the sources linked above and take advice on your own circumstances from a lawyer qualified where you are.
Read next
Generative AI guidance for lawyers: how six markets compare
How to find what regulators, courts and bar bodies have said in six markets, and the common themes across them.
5 minute readEU AI Act: what it means for firms outside the EU with EU clients
How a firm outside the EU should think about the EU AI Act when it works for European clients.
5 minute readUK: data protection law when AI processes client data
Controller and processor roles, lawful basis, transfers and the records a UK firm should keep when using AI services.
5 minute read