The first SRA-approved AI law firm: what it means

On 6 May 2025 the Solicitors Regulation Authority announced that it had authorised Garfield.Law Ltd, the first purely AI-based law firm allowed to provide regulated legal services in England and Wales. It offers an AI litigation assistant for recovering unpaid debts through the small claims process. The SRA checked quality control, confidentiality, conflicts and hallucination risk, and named solicitors remain accountable.

The announcement was widely reported, and some headlines read more into it than the SRA's release does. This piece sticks to what the SRA's own release says, then draws out what it means for firms that have no intention of becoming AI-driven but are using, or thinking about using, AI somewhere in the practice. It sits in our guide to AI for UK law firms.

Which law firm did the SRA approve as AI-driven?

Garfield.Law Ltd. According to the SRA, it offers small and medium-sized businesses, as well as law and other firms, an AI-powered litigation assistant to help them recover unpaid debts, guiding them through the small claims court process up to trial. The SRA described it as the first purely AI-based firm it has authorised, while noting that many firms already use AI to support back-office and public-facing services.

That distinction matters. Using AI inside a conventional firm did not need, and does not need, special permission. What was new was a firm whose service is delivered through AI.

What did the SRA check before approving an AI law firm?

The SRA said it engaged with the owners to consider the firm's processes and assure itself that its rules can be met by an AI service. Its release lists what it sought reassurance on:

• Processes to quality-check work.
• Keeping client information confidential.
• Safeguarding against conflicts of interest.
• Managing the risk of 'AI hallucinations'.

It also set out three features of how the system works that bear directly on those risks:

• The system will not be able to propose relevant case law, which the SRA called a high-risk area for large language models.
• It is not autonomous. It will only take a step where the client has approved it.
• There are supervision and monitoring processes, including greater oversight of claims in the initial launch phase so that issues can be identified.

Who is accountable when an AI law firm gets it wrong?

People. The SRA's release says that under its rules, named regulated solicitors remain ultimately accountable for the firm delivering high professional standards, and that they are responsible for all the system's outputs and for anything that goes wrong. It also notes that all regulated firms must hold a minimum level of insurance to protect clients.

This is the single most important line in the announcement for everyone else. Being authorised to use AI did not move responsibility onto the software, the vendor or the model. If that is true for a firm built on AI, it is certainly true for a firm that uses an AI tool in one corner of its work.

Why did the SRA approve an AI law firm?

The SRA's stated reason is consumer benefit. It said AI-driven services could deliver better, quicker and more affordable legal services, and its chief executive, Paul Philip, linked the decision to the number of people and small businesses struggling to access legal services. The release also says the risks around an AI-driven firm are novel, that the SRA worked closely with the firm to make sure it could meet the rules, and that it expects this to be the first of many such firms, whose progress it will monitor closely.

That is consistent with the SRA's wider position. Its Risk Outlook on AI says it regulates the outcomes firms achieve rather than the specific systems they use, and that it does not recommend individual providers or software.

Does this mean AI can give legal advice now?

No. Garfield is a regulated firm, with solicitors accountable for its work, offering a narrowly defined service in debt recovery. The approval did not create a general permission for AI tools to advise the public, and it did not change what the Code of Conduct requires of any firm. A chatbot on an unregulated website giving legal opinions is in a completely different position from an authorised firm operating under supervision.

For firms deploying client-facing AI, the lesson points the other way. The regulator was comfortable with a system that cannot cite case law and that acts only with approval. Those are constraints, not capabilities. Our piece on what AI chatbots can safely say applies the same thinking to the front door.

What does the SRA's decision mean for other law firms?

Five lessons, drawn from what the SRA said it checked. None of them requires your firm to be AI-driven.

• Narrow the job. Garfield does one defined thing. A tool that does one job well is easier to supervise than one that does everything a little.
• Remove the riskiest capability rather than managing it. The system cannot propose case law at all. If a tool in your firm does not need to produce authorities, make sure it cannot.
• Put a person at each consequential step. Garfield acts only with the client's approval. Decide which actions your tools may take on their own, and keep that list short. See AI agents for law firms.
• Supervise most at the start. The SRA noted extra oversight in the launch phase. Read every output for the first weeks of any new tool, then sample.
• Name who is accountable. Named solicitors own the outputs. Your AI policy should name a partner, not a department. AI governance for law firms has a one-page template.

A self-check: could your AI use pass the same questions?

Take the four things the SRA sought reassurance on and ask them of your own firm's AI use, tool by tool. A 'no' does not mean you are in breach of anything. It means you have found the gap to close first, before anyone else finds it for you.

• Quality: for each tool, can you name who checks its output and show that they do?
• Confidentiality: do you know where each tool sends client information, and has the firm agreed to it?
• Conflicts: could a tool that talks to the public end up speaking to both sides of a matter, and what would it do if it did?
• Hallucinations: is there any route by which a made-up fact, date or authority could reach a client or court without a person checking it?

A worked example, invented for illustration. A hypothetical firm runs the self-check and finds one clear gap: an AI assistant a trainee uses for research notes, with nobody checking the citations before they reach a fee-earner. The fix costs nothing: the fee-earner who relies on a note checks every authority in it, as the Divisional Court said in Ayinde they must. AI hallucinations and law firms sets out the routine.

What is the difference between an AI law firm and a firm that uses AI?

It helps to separate three positions that are often run together in coverage of this story.

• A firm that uses AI tools. Most firms, in some form. A solicitor drafts with help from a general assistant, or the firm's phone is answered out of hours by an AI that takes details and books a call back. The legal service is still delivered by people. No special authorisation is involved.
• An AI-driven firm. The legal service itself is delivered through an AI system, with solicitors supervising and accountable. This is what the SRA authorised in Garfield's case, after examining how the firm would meet its rules.
• An unregulated AI tool offering legal help to the public. Not a law firm at all, not authorised by the SRA, and without the protections that come with instructing a regulated firm, such as the insurance the SRA mentioned.

Most of the practical questions partners ask us sit in the first category. For them, the Garfield decision is useful less as a precedent to follow than as a description of what the regulator looks for. A firm using AI to answer its phone is not trying to become Garfield, but it can borrow the same disciplines: a narrow job, no case law, a person at each consequential step, closer supervision at the start and a named solicitor who owns the outcome.

If you are a business considering an AI-driven legal service, the SRA's register of authorised firms is the place to confirm that a provider is regulated. If it is not on the register, the protections the SRA described do not apply.

Will there be more SRA-approved AI law firms?

The SRA said it expects Garfield to be the first of many, and that it has a team working with firms and technology providers to test new ideas. We are not in a position to predict which firms or when, and would treat any list of 'the next AI law firms' with caution unless it links to an SRA announcement. The register of authorised firms, on the SRA's own site, is the place to check whether any firm is regulated.

What this means for the front door

We are not an AI law firm and do not provide legal services. We build for the front of a conventional firm, and we find the SRA's constraints a useful design brief. Cam, our AI Engagement Coordinator, takes calls on a firm's line at any hour, says it is AI, asks the firm's own qualifying questions, books the call back and sends the firm a summary. It does not cite case law or give advice, and the solicitor decides what happens next. For the regulatory side of AI on the phone, see can a UK law firm let AI answer client calls. If you would like to talk it through with your COLP, we are happy to.

Frequently asked questions

What is the first SRA-approved AI law firm? Garfield.Law Ltd, authorised by the SRA and announced on 6 May 2025. It offers an AI-powered litigation assistant to help businesses and firms recover unpaid debts through the small claims process.

Can Garfield.Law's AI cite case law? No. The SRA said the system will not be able to propose relevant case law, which it described as a high-risk area for large language models.

Who is responsible if an AI law firm makes a mistake? The SRA said named regulated solicitors remain accountable for all the system's outputs and for anything that goes wrong, and the firm must carry the minimum required insurance.

Do ordinary law firms need SRA approval to use AI? The SRA says it regulates outcomes rather than specific technologies and does not approve individual software. Firms using AI remain subject to the same rules as for any other work. Check specifics with your COLP.