The doorway matters less than who answers for what happens behind it
The Law Society’s foresight report, published on 5 October, offers a memorable image: agentic AI as the “default doorway” into the justice system, spotting a person’s legal problem, explaining their rights and running the process for them. Our news report covered the headline findings. The more consequential argument sits further into the 88-page document. If that doorway is run by commercial platforms, someone’s route to a remedy is shaped by business models and design choices rather than in a courtroom. The report’s answer is uncomfortable for every institution involved: governance as currently built arrives too late.
Strategic Reality: The report does not claim agentic AI is in courts today. Its own summary says fully agentic AI “is not yet used in the justice system but looks likely to emerge”. The warning is about the order in which decisions are being made, not about a technology already deployed.
Why the doorway is already being built
Consumer-facing AI in legal disputes is not hypothetical. JUSTICE surveyed 3,287 people between December 2025 and January 2026. Of the 1,428 who had been in a legal dispute during the previous two years, 233, or 16.3%, had asked an AI chatbot for guidance on it. We covered that research when it appeared. Chatbot users were also less inclined to have gone to a solicitor: 33.9%, against 44.6% of people who did not use one.
Law Society members consulted for the report described the same pattern. One said: “Litigants in person are all using AI anyway.” The speaker added that much of what gets produced is “nonsense” and is “creating more work within the system”.
Today’s chatbots answer questions. Agents take actions: drafting the claim, sending the correspondence, choosing the next step. The report frames the shift as a move from human decisions that machines assist to machine actions that humans supervise.
| Figure | What it measures | Why it matters |
|---|---|---|
| 16.3% | Share of people with a recent legal dispute who asked an AI chatbot (JUSTICE) | A consumer route into legal help already exists outside the profession |
| 33.9% vs 44.6% | Chatbot users vs non-users who sought a solicitor’s advice (JUSTICE) | The people using the new route are consulting solicitors less often |
| 19 + 11 | Interviews behind the Law Society’s findings | 19 came from earlier work on agentic AI in legal practice; 11 more were with justice-system experts |
| 5 | Case types the Law Society names as suited to agentic tools | Probate, debt recovery, small claims, road traffic penalties and employment tribunal claims |
Critical Context: Legal Futures summarised the method as 19 interviews plus 11 with justice experts. The report itself is more precise: the 19 conversations fed the Law Society’s earlier report on agentic AI in legal practice, and this study added 11 focused interviews, desk research across China, Estonia, the UK and the US, and member discussions.
Where does accountability go when the agent sits outside the court?
Responsibility is spread across people who never meet the case
Professional regulation is built on the idea that a person answers for the work. The report states the problem plainly: interviewees found that “legal regulators regulate people, not technology”. Once software does work that only a regulated professional could otherwise do, that assumption breaks.
The report’s alternative locus is upstream. In its account, responsibility ends up divided between the companies that build the tools, the courts that run them, the datasets they rely on, and whoever designed and bought them. Its suggestion is that frameworks “may need to be adapted” to reach buying decisions, product design and default settings, all of which happen earlier than today’s oversight engages.
Hidden Cost: The report observes that, without deliberate choices about responsibility, institutions would capture the efficiency gains while the cost of errors could land on individuals who are “least likely to recognise that a mistake has been made, least equipped to challenge it, and least able to seek redress”.
The market may set the boundaries first
The report warns that when consumer appetite and commercial reward drive uptake, tools can become entrenched before anyone understands what they do to outcomes. Regulators and courts could then find themselves approving whatever has already caught on, instead of influencing how it was built.
The report also flags a quieter commercial risk. Uncertainty over whether a developer, a deploying court authority or a commissioning agency is liable for a wrong administrative decision “may constrain the appetite to scale”, and insurers “may be reluctant to underwrite” that exposure while it stays unclear. Unresolved liability could slow adoption as well as expose users.
”Routine” is a judgement, not a category
The report expects agentic tools to arrive first in support work and highly routine cases, and then asks the obvious question: who decides what is routine? Labelling a case straightforward enough for automation is a judgement in its own right, involving risk, vulnerability and circumstance. A file may appear simple precisely because context is absent, or because the system stripped out what made it hard. The report’s conclusion is that delegation may grow subtly, without any explicit transfer of authority.
Reality Check: The report’s most credible early uses are administrative: processing documents, sorting and listing cases, and carrying out orders once a decision is made. It calls this “one of the few areas” where the technology “aligns cleanly with system need”. On our reading, sorting cases is also where the “routine” judgement gets made, so the safest use case and the most important classification decision sit in the same place.
Who is affected
| Group | How the report says they are affected |
|---|---|
| Solicitors and law firms | The role may shift towards validating, correcting or reframing AI-driven decisions rather than guiding clients through legal principles |
| Judges and court administrators | The judge still signs, but the analysis behind a judgment may be assembled by a system; reliance on AI could reduce the practice judges get in the work they supervise |
| People without digital access, confidence or literacy | If agents become the main entry point, lacking digital skills or devices could mean losing access to the courts altogether |
| Better-resourced users | Earlier access to better commercial tools, which the report says could create unequal legal help long before anyone formally acknowledges it |
| Regulators | Agentic systems may have no fixed decision points to target, and responsibility is hard to assign across design, data and user input |
The Law Society’s summary page puts the inequality point directly. Getting value from an agent requires the user to be online, confident and digitally literate. Wealthier users “will also benefit from earlier access to higher-quality tools through commercial providers”.
What this means for UK law firms
The report does not treat the profession as a bystander. One member put the existing position bluntly: “Lawyers are responsible for their work.” The report records some members’ view that deploying a tool you cannot competently check already falls short of a solicitor’s existing duties. That matches the supervision duty we examined in our analysis of AI hallucination as a supervision problem.
The structural change is in what clients bring through the door. If agents become people’s first stop, the report suggests solicitors may spend more time checking, correcting or reframing what software has already done. The report expects solicitors’ value to rise as the people able to contest an automated route and defend fair procedure. Interviewees also asked how future lawyers will learn the job if the junior work that once trained them is automated away.
Strategic Insight: Law Society vice-president Brett Dixon framed the issue as a choice about limits: “There is still time to shape the boundaries of AI before the market does it for us.” The report itself says the Law Society’s role is shifting “from guidance to boundary-setting”.
Open challenges
What counts as agentic. The report suggests the definition itself could turn into a regulatory fight. The riskiest systems may not be the ones labelled autonomous, but those that reach similar effects by combining smaller capabilities. A definition drawn too narrowly lets responsibility slip through it.
Whether human oversight stays meaningful. The interviews produced no consensus on the right level of human checking, or on its form. The closest thing to agreement was that anyone affected by an automated decision should be able to request a human review. The report warns that a safeguard can survive on paper while disappearing in use, when a system is too intricate, too repetitive or simply too easy to wave through.
Systems that cross agencies. An agent that triggers actions in several organisations’ systems falls between regimes built to regulate one system at a time. The report argues these deployments must be judged as one joined-up whole.
Efficiency against rights. The dominant view among interviewees was that faster, cheaper justice can itself protect rights. A minority argued the burden of proof lies with those claiming rights are adequately protected, and that it has not been met.
⚠️ Warning: The report’s own forecast for how this gets tested is not reassuring. It suggests that the first real test of governance “may be a high-impact failure” rather than a well-planned framework.
What to watch
The Online Procedure Rule Committee. The report names the OPRC as an existing body that could shape procedural standards should agents be used to sort, manage or resolve disputes online. Its first draft rules went out to consultation in December 2025, starting with possession proceedings. If the final rules address AI-prepared or AI-submitted filings, procedure is starting to absorb agentic tools. If they are silent, the doorway will be shaped by products rather than rules.
Consumer protection as the first regulatory lever. The report cites guidance on agentic AI from the Competition and Markets Authority, issued in March 2026, which confirms that consumer protection law already covers these systems while recognising risks that law was never written for. Enforcement action against a consumer legal agent would show whether consumer law, rather than legal services regulation, becomes the practical constraint.
The Law Society’s boundary list. According to the report, interviewees want the Law Society to help settle three things: which judgements must never be handed to a machine, what real oversight looks like, and which protections must exist before software can act on someone’s behalf. A published position on those three questions would turn the report’s foresight into something firms and providers can be measured against.
Who offers a human route. Interviewees insisted that any service designed around AI first must still offer people a real human option. Watch whether new court and dispute services guarantee one, or treat it as an exception.
The report ends by asking what a justice system should “remain human for”. For UK firms and courts, the practical form of that question is narrower. Every procurement and design decision taken now is an answer to it, whether or not anyone frames it that way.
Source: Agentic AI could become “default doorway” into justice system (Legal Futures, 6 October 2026), by Nick Hilborne. Additional sources: The future of agentic AI in the justice system (The Law Society, 2026), including the full report and summary; What AI Chatbots Can Teach Us About Unmet Legal Needs (JUSTICE, 2026).
This strategic analysis was written by Resultsense, a UK-focused AI news and analysis publication. We will be watching the Online Procedure Rule Committee’s final rules and the Law Society’s next move on non-delegable functions for the first sign of where the boundaries will actually sit. Read more analysis at Insights, or get in touch.