The Court of Appeal handed down its decision in CILEX v Mazur and others ([2026] EWCA Civ 369) in March, and the Solicitors Regulation Authority followed with revised supervision guidance on 12 June. Read together, they tell a firm a good deal about delegating work to people. They tell it less about software conducting the work of litigation, and that gap is the point the Law Society has now asked the regulator to close.
What Mazur settled
The case turned on who is allowed to conduct litigation, a reserved activity under the Legal Services Act 2007. The Court of Appeal held that a person who is not themselves authorised can still perform tasks within the conduct of litigation, provided an authorised individual keeps proper direction, management, supervision and control throughout. The earlier High Court view had been stricter. The practical effect is that paralegals, legal executives and trainees can carry real litigation work, so long as a named authorised person stands behind it and remains accountable.
The SRA turned that principle into its updated guidance, produced in consultation with the Law Society, CILEX Regulation, the Legal Aid Agency and the Law Centres Network. It sets no single model of supervision. Instead it asks each firm to calibrate oversight to the nature of the work, the experience of the person doing it, and the harm to the client if the work is defective.
The question it left open
Mazur was argued about people. Its logic runs straight into a question about machines. If an unauthorised person needs direction and control to work on a piece of litigation, what is the position when the actor is an AI tool that drafts a statement of case, decides which authorities to cite or settles the terms of an application? The Law Society has put the point plainly, saying the legitimacy of using AI to make key decisions in a case that would amount to conducting litigation if taken by an individual remains unresolved. It has asked the SRA to give clear guidance on what is permissible as a matter of urgency, so that firms are not held back from testing sensible uses of these tools while the position is uncertain.
That uncertainty carries weight because conducting litigation without authorisation is no minor breach. It is the kind of thing that brings regulatory and, in some circumstances, criminal consequences. A tool that quietly moves from assisting a lawyer to making the decisions that define a case can put a firm on the wrong side of that line without anyone intending it.
Why it reaches your firm
A smaller firm might read all this as a debate for large litigation practices. It is not. The supervision principle in Mazur applies to every SRA-regulated firm, and the AI question applies to any firm that lets a tool touch contentious work. If a fee earner uses an AI system to prepare correspondence in a dispute, produce a witness statement or decide the shape of an application, the firm needs to be able to say that an authorised person directed and checked that work. Deferring to the machine is not direction. The distinction the court drew between doing the work and controlling the work is the same distinction a firm has to hold onto when the worker is software.
What a firm should do
Treat AI in contentious matters as delegated work, not as a shortcut that sits outside supervision. Name the authorised person responsible for each file and make clear that they own the decisions, not the tool. Record where AI is used in litigation and what a human checked before anything left the firm. Watch for the SRA guidance the Law Society has requested, because it will draw the boundary the profession is currently guessing at. Until it arrives, the safe reading is the strict one. An AI tool assists the lawyer, and the lawyer, not the tool, conducts the litigation.
The revised supervision guidance is in the SRA's guidance library.
If your firm delegates litigation work to software in any form, the supervision question in Mazur is yours to answer, and we help firms answer it in writing: start with a conversation.