Following on from his regular series on AI in conveyancing, Ed Molyneux adopts a different approach. This week, he begins to explore the safe adoption of AI. In the first of a new series, he reads the SRA’s new warning notice as a design brief, and sets out how a firm builds an environment in which the regulator’s line is hard to cross.
On 17th August the Solicitors Regulation Authority (SRA) published a warning notice on the misuse of AI. Its message is blunt. Using AI “does not reduce, remove or replace the professional, regulatory or legal responsibility” a solicitor bears, and “reliance on an output of AI would not be a suitable defence” when something goes wrong.
A warning notice tells you what not to do. It does not tell you how to build a practice in which the wrong thing is hard to do in the first place. That gap, between the regulator’s line and the working arrangement that keeps you the right side of it, is the subject of this series. This opening article maps the notice onto the build.
What the notice actually warns about
The warning notice names two failure modes, both drawn from real cases. The first is fabricated authority: AI producing case citations or references that look genuine and are not, then reaching a court. The notice is unsparing. A solicitor must “be assured that all submissions of named case law authorities are genuine, relevant, have a verifiable citation”, and cannot hide behind the tool.
The second is breach of confidentiality: entering client information into AI systems where, depending on the provider’s terms, it “may be stored, retained or used to improve the tool”. The notice is careful not to confine this to free tools, warning that “both free to use and paid for AI systems may pose risks to client confidentiality”. It is starkest with public systems, where uploading privileged material can place it in the public domain, with the waiver of privilege that implies.
Around these, the notice restates duties that were always there: the Principles, competence and supervision, confidentiality under Code 6.3, and the firm’s own accountability for work carried out through others. None of this is new law. What is new is the regulator saying, in terms, that AI changes none of it.
Reading it as a conveyancer
The fabricated-authority risk is most vivid in litigation, and conveyancers are not filing fake case law every day. But the notice binds every solicitor, and its confidentiality, supervision and accountability points land squarely on conveyancing, a practice that handles large volumes of personal and financial client data and is adopting AI quickly.
Its data requirements, in particular, read almost as a specification. A firm using AI with client data must be able to confirm the data remains within a secure environment, is not accessed by unauthorised third parties, is not used to train AI models except where explicitly authorised, and is not retained longer than necessary. Read that list again and notice what it is not. It is not “do not use AI”. It is a set of properties your AI arrangement has to have, and be able to evidence. That is a build brief.
The notice is the ‘what’. The rest is the ‘how’
Each thing the SRA requires has a corresponding move. A secure environment, not accessed by unauthorised third parties, is an access-control property, and you cannot get it by asking staff to be careful. You get it by giving AI a single, governed route to client data instead of a paste box, a route that establishes who is asking, scopes what they may see, and lets nothing reach an unapproved tool.
The four data requirements, security, unauthorised access, training and retention, are precisely the questions a firm must be able to answer about any AI supplier, because “we are secure” is not one of the confirmations the SRA asks you to obtain. And supervision means remaining accountable for the work: you can only supervise what you can see, so an arrangement that records who used AI, on which matter, and what data it touched is what makes accountability evidenced rather than merely asserted.
That leaves the hardest line of all. “Reliance on an output of AI would not be a suitable defence” is the accuracy duty, and it is a different axis from confidentiality entirely. Governing where data goes says nothing about whether the answer is right, and that is a question this series will close on.
The takeaway
The SRA has not banned AI, and no one should read it that way.
It has said, plainly, that the responsibility is yours whatever tool you use, and it has listed the properties a safe arrangement must have. The firms that come out of this well will treat that list not as a threat but as a design brief, and build the environment rather than issue the memo.
What that environment looks like, from first principles, is what the rest of this series sets out.
This is the opening article in a series on adopting AI in a way a conveyancing firm can genuinely stand behind. This article is general information, not legal advice.
About the author
Ed Molyneux is co-founder and CTO of Moverly and the original author of the Property Data Trust Framework (PDTF), the open standard for machine-readable property data now being adopted across the industry. Ed writes about AI, property data infrastructure, and the future of conveyancing.
