Posted by Yazad Bajina, chief commercial officer at Legal Futures Associate Kord
Baijina: Do not wait to retrofit compliance onto agents
Everyone’s talking about agentic AI.
The Solicitors Regulation Authority (SRA) is actively exploring the benefits and challenges of embedding it in legal services, but the Law Society’s April 2026 foresight report made the somewhat awkward finding that there is currently no evidence that agentic AI is actually being used in legal practice.
So, does that mean the legal profession is wasting its breath on something that’s not even being used?
Not at all. There is a sense that agentic AI adoption is inevitable, and it’s worth the profession thinking about what that will mean now, rather than after it’s already ubiquitous.
Remaining accountable
Among the challenges posed by any future adoption is that solicitors remain personally accountable for every piece of work, regardless of whether AI was involved.
Using an AI tool does not, by the letter of the law, transfer or dilute professional responsibility.
But AI agents can complete multi-step tasks with limited human oversight. That may mean the solicitor is not meaningfully in the loop when their digital employee gets to work.
So, while the assumption remains that the solicitor gets final sign-off on whatever the AI agent does, can they be truly accountable when they may not even be aware what the agent is doing?
Active supervision
An AI agent is difficult to supervise. By design, it works in the background, beyond the reach of ordinary human review. But that does not remove the accountability; it simply means a solicitor can be held responsible for work they had no real means to oversee.
Responsibility you have no way to exercise is not accountability at all.
A solution of sorts has come from the SRA itself.
To perform the supervisor’s role effectively, it suggests, the accountable person needs sufficient information, such as data logs that let them reconstruct what the agent has actually been doing.
Genuine accountability, in other words, depends on the conditions for oversight being in place first.
The courts are heading in the same direction. In the well-known case of Ayinde, the Divisional Court held last year that a tool such as ChatGPT was not capable of conducting reliable legal research and warned that lawyers who put unchecked AI output before a court risked sanctions as serious as contempt of court proceedings.
The message for firms is not simply that individuals must take more care, but that wherever AI is in use, the firm needs to put in place structures for verification that can show how an output was produced and checked.
But for the overwhelming majority of firms these conditions simply don’t exist. And given the slow pace of adoption so far, why would they?
The time is now
The reality is that firms will still be held accountable for what their AI agents do. Even if solicitors themselves can’t realistically oversee what’s going on behind the AI interface, they still need to establish the right conditions, structures and processes to ensure outputs are accurate.
That means the time to get those structures in place is now, before workflows start to depend on AI agents.
If firms wait until agents are already well established, they’ll be retrofitting compliance and oversight into processes already running. And that’s a much harder prospect.
Leave a Comment