In the first of a new series on safely adopting AI, Ed Molyneux began to explore the topic using the SRA’s new warning notice as a design brief and set out how a firm builds an environment in which the regulator’s line is hard to cross. In the second instalment, he explains why the tool a firm chooses is the last question, not the first, and how starting instead from the duties a practice already holds makes the right approach more or less design itself.
Almost every conversation about AI in a conveyancing firm starts in the same place: which tool should we use? Which model, which vendor, which features?
It is the last question, not the first. The duties that govern AI use are duties a firm already holds, whether or not AI is involved. The SRA made this plainly in its August 2026 warning notice, stating that using AI does not reduce, remove or replace a solicitor’s responsibility. Get those duties clear and the right way to work more or less designs itself. Start with the tool, and you end up with a drawer full of vendor assurances and no way of knowing whether they add up to anything.
So it is worth setting the tools aside and starting from the obligations. There are five that any use of AI in a conveyancing practice has to satisfy, and stating them plainly is the whole foundation.
Start with accountability
Accountability comes first, and it is the one most AI mishaps actually breach. Data protection law requires a firm not merely to comply, but to be able to demonstrate that it has.
That is a higher bar than doing the right thing in fact. A practice can hold a perfectly sound lawful basis, act in good faith and still fall short here, because when asked “what did your AI use touch last month, and can you show it stayed within approved bounds?”, it cannot answer.
Most AI failures in professional practice are not failures of lawfulness – they are failures of demonstrability.
Risk-appropriate safeguards
Second is security. The law expects measures appropriate to the risk, and for a firm holding confidential client files that is not an abstraction.
It sensibly means knowing who accessed what, controlling which systems client data is allowed to reach, and being able to notice when it reaches somewhere it should not. An AI tool is simply another system that data can flow into; the question is whether that flow is controlled and visible, or neither.
The SRA puts it in its own terms: client data must stay within a secure environment and not be accessed by unauthorised third parties.
A lawful basis
Third is a lawful basis, and here a common misconception does real damage. A firm does need a lawful basis to process client personal data, in practice the performance of the retainer or its legitimate interests.
But that basis is a determination made once, for a purpose, and recorded. It is not minted afresh every time a fee-earner opens a tool. Treating each use of AI as a new lawful-basis event is wasted effort, and it distracts from the obligations that genuinely bite.
Fourth is processor governance. Where a third party processes personal data on the firm’s behalf, a written agreement must govern it. Read forwards, that is a design constraint: client data may flow only to processors the firm has actually assessed and contracted with.
An AI vendor is a processor. So which tools may receive client data is not a matter for individual discretion under deadline pressure. It is a controlled list, and something has to hold the line on it.
Professional duty
Fifth is professional duty. Beyond data protection, the regulatory frameworks impose obligations that AI cannot be allowed to erode: keeping client affairs confidential, acting with competence, and, the one that matters most as these tools improve, not surrendering professional judgement to a machine.
The SRA is explicit that reliance on an AI output would not be a suitable defence. A tool may support a fee-earner’s assessment. It may not quietly become it. And a firm remains responsible for supervising how its people use these tools, which it cannot do if it cannot see what they are doing.
Read those five together and they stop being a compliance checklist and become something more useful: a specification. Each one implies something the firm’s set-up has to provide.
Implied meaning
Accountability implies a record of what AI use actually touched. Security implies controlled, visible data flows tied to identity. Lawful basis implies a decision made once and applied consistently, not pushed down to the fee-earner at the point of use.
Processor governance implies enforcement that data reaches only approved vendors. Professional duty implies AI positioned as support, with a human in the loop and a supervisor able to see the work.
Notice what none of these is satisfied by: a better prompt, a sterner policy, or another training session. They are properties of the environment the AI runs in: how people sign in, what the tool is allowed to reach, and whether every use leaves a trace. That environment is a thing a firm can design deliberately, and it is what the rest of this series sets out to build.
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.
The views expressed in this article are those of the author and not necessarily those of Today’s Conveyancer. This article is general information, not legal advice.
