Lawyers, the Algorithm, and the New Standard of Care

The legal profession has always been adept at turning social change into doctrine. It takes a new technology, a fresh risk, a novel commercial habit, and then quietly, sometimes without fanfare, converts it into an expectation. What begins as optional convenience becomes best practice; what begins as best practice hardens into duty. This week, that familiar alchemy is being attempted again, this time with artificial intelligence, and with unusually blunt language about the personal consequences for those who decline to keep up.

Guidance issued by the UK Jurisdiction Taskforce, a body backed by the Ministry of Justice and chaired by Sir Geoffrey Vos, the Master of the Rolls, warns that lawyers could find themselves on the wrong end of professional discipline, including being struck off, if they fail to use AI where it would materially improve their work. The point is not that every solicitor must subscribe to the same software, or that every barrister must delegate their submissions to a chatbot. It is, rather, a claim about standards: that the “reasonable care and skill” expected of a competent professional may soon include, in some tasks, the competent use of machine assistance.

In plain terms, the Taskforce is sketching a future in which refusing to use AI is not a quaint personal preference but a potential breach. Its statement says professionals should be alive to the possibility that they could be liable for failing to use AI for a task when a professional exercising reasonable care and skill would have done so. Matthew Lavy KC, a barrister who sits on the Taskforce panel, put it even more starkly: we are not far, he said, from a time where harm caused by a failure to use AI, where a reasonable lawyer would have done so, will be actionable.

This is not merely a nudge towards modernisation. It is an attempt to redraw the baseline of competence. For centuries, the law has judged lawyers against their peers, and the peer group has tended to move at the pace of tradition. Now the suggestion is that the peer group itself will be pulled forwards by tools that promise speed, pattern recognition, and relentless consistency. The Taskforce even offered an analogy designed to sting: a radiologist who refuses to use AI tools that are extremely effective at identifying cancerous tumours. The implication is clear. A professional who knowingly ignores a demonstrably helpful technology might not be defending principle; they might be courting avoidable harm.

Set out like this, the argument has a certain practical appeal. Clients do not pay for nostalgia. They pay for outcomes: a contract that does not unravel, a claim that does not collapse, a merger that does not stumble on an obvious regulatory snag, a disclosure exercise that does not miss the document that mattered. If AI can truly reduce error rates, widen the sweep of review, and allow lawyers to spend more time on judgement rather than drudgery, then there is an obvious public interest in encouraging adoption. The law is a service industry with unusually high stakes, and its mistakes are not theoretical. They are measured in prison sentences, financial ruin, and families that never quite recover.

But the guidance arrives with an irony that the Taskforce itself acknowledges. The same technology that may become professionally obligatory is also capable of producing precisely the kind of error that courts and regulators detest: confident nonsense. The warning is therefore two-pronged. Lawyers might be punished for refusing to use AI where it helps, but they might also be punished for misusing it, including where the technology is responsible for mistakes. If that sounds like an impossible tightrope, it is because the profession is being asked to master a tool whose strengths and weaknesses are not yet widely understood by the public, and not always admitted by the vendors selling it.

Recent cases have made the risks difficult to dismiss. The Taskforce’s statement follows a string of episodes in which lawyers submitted flawed documents containing AI-induced errors, including false citations and fictitious cases. In one high-profile incident, Pinsent Masons, one of Britain’s biggest law firms, was criticised by a judge after AI “hallucinations” appeared in court filings. The firm apologised and referred itself to the Solicitors Regulation Authority. Few episodes have done more to puncture the mystique of AI competence than a legal submission that looks plausible until one checks the authorities and finds they never existed.

This is the new paradox of professional AI: it can be dazzling at summarising and drafting, and dreadful at knowing when it is wrong. If a junior lawyer makes up a case, a supervisor can usually read the paragraph and sense that something is off. If a model invents an authority with a plausible name, and sprinkles it among genuine references, it can slip through hurried review, especially in a system already under pressure from cost-cutting and time billing. The danger is not just that AI can make mistakes, but that it can make them at scale, quickly, and with a kind of linguistic confidence that has long been mistaken for competence in legal writing.

Then there is confidentiality, the oldest and most brittle pillar of legal practice. The Taskforce’s guidance notes that lawyers could also be punished for putting confidential legal documents into AI systems that fail to protect client privacy. The warning aligns with the Bar Standards Board’s own caution, issued in May, that serious sanctions could follow if lawyers feed privileged material into publicly accessible AI systems such as ChatGPT. Here, the legal profession is not merely policing quality; it is policing trust. Once clients suspect that sensitive documents are being poured into general-purpose online services, potentially retained, learned from, or exposed through future breaches, the lawyer’s promise of discretion begins to look fragile.

This, too, has a practical dimension. Much of the current consumer market in AI is built on convenience. People paste text into a prompt box and accept the answer. Many legal tasks, however, involve precisely the information that cannot be pasted anywhere: witness statements, medical reports, commercial strategy, the internal messages of a company under investigation. If the profession is to adopt AI widely, it cannot be the casual, consumer-grade sort. It must be secure, auditable, and designed around duties of confidentiality rather than around speed to market. That is a procurement problem, a governance problem, and, increasingly, a regulatory problem.

It is also, quietly, an access to justice problem. Large commercial firms can buy bespoke systems, negotiate enterprise contracts, and build internal policies. Small practices, legal aid providers, and self-employed barristers may be offered a different reality: a patchwork of tools, uncertain training, and financial pressure to do more with less. If regulators and senior figures in the system begin to treat AI use as part of reasonable competence, they will have to confront the question of what happens to those who cannot afford the best tools, or who are working in corners of the market where margins are already thin.

The answer might be that AI, by reducing the cost of certain tasks, will in fact strengthen the smaller end of the profession. There is a plausible story here: the sole practitioner able to produce a decent first draft in minutes, the overstretched employment solicitor able to triage cases more quickly, the junior who can interrogate a bundle with more thoroughness than their exhausted eyes would allow at midnight. There is also a plausible darker story: a race to the bottom in which clients expect lower fees because “the machine did it”, while lawyers carry the same liability for the output and face discipline when the tool misfires.

That liability question is the fulcrum. Professional duty has never been purely about effort; it is about meeting a standard. If AI demonstrably improves performance in certain tasks, and if a competent practitioner in future would be expected to use it, then refusing to do so could look like negligence. But the law will not give lawyers a free pass when the machine errs. The Taskforce’s stance suggests an emerging doctrine that might be summarised as: use it, but do not trust it. That is sensible in theory. In practice, it may be the most difficult professional habit to cultivate, particularly in a culture that has long equated fluent prose with sound reasoning.

We can see the beginnings of the new etiquette. AI might be used to generate a first draft, but every citation must be checked. AI might be used to summarise evidence, but the underlying documents must be sampled and verified. AI might be used to search for patterns in disclosure, but the selection criteria must be recorded. AI might be used to propose arguments, but counsel must test them against the facts and the law. In other words, AI becomes a tool for acceleration, while responsibility remains stubbornly human. Regulators are unlikely to accept the defence that a lawyer did not know what the system was doing, or that the output looked convincing at the time.

That brings us to the psychological shift. The profession is used to delegating. Juniors draft; seniors review. Paralegals search; associates sign off. But delegation works because the delegate is another person with a career, a training path, and an ability to explain their reasoning. AI does not have those features. It cannot be cross-examined. It cannot be disciplined. It cannot, in any meaningful sense, be accountable. A lawyer who uses AI is therefore not delegating responsibility; they are introducing an opaque instrument into their workflow and accepting that they remain the sole address for blame.

The guidance is also part of a broader cultural struggle over the future shape of the profession. For some senior figures, AI promises a more efficient system, one in which routine work is automated and lawyers spend more time on judgement, negotiation, and advocacy. For others, it threatens to erode the apprenticeship model that has long trained lawyers through repetition of the supposedly mundane. If drafting becomes instantaneous, where will the novice learn the structure of an argument? If research becomes conversational, how will the trainee develop the instinct for what matters? A profession can automate itself into competence problems if it removes too many opportunities for learning.

And yet, the pressure to change is now being applied from several directions at once. Courts are tired of sloppy filings. Clients are tired of expensive hours spent on tasks that look mechanical. Competitors are offering faster turnaround. Regulators, perhaps anticipating public anger at avoidable mistakes, are beginning to signal that the old defences will not hold. The Taskforce’s intervention is notable because it frames AI not as an optional competitive advantage but as a component of the baseline that protects the public.

It is not hard to imagine how this will play out in disputes. A negligence claim may soon feature expert evidence about prevailing AI practice: what tools are used for document review, what verification steps are standard, what training is typical, what audit trails are expected. Litigation will acquire a second shadow litigation about process. Did the lawyer use an AI system? Which one? Did they check the output? How? Was the prompt stored? Were confidential documents uploaded to a public interface? Was there a policy? Was it followed? These questions will become as routine as questions about supervision, attendance notes, and conflicts checks.

The most striking illustration of where this could lead is not in a law firm but in the emergence of automated legal services. Last month, a completely automated AI law firm, Garfield AI, won a case in court for the first time in British legal history, acting on behalf of an HR professional seeking £7,000 from a former employer. The symbolism is powerful. It suggests that the boundary between software and representation is already being tested in the open. For regulators, it raises a question that can no longer be postponed: if an automated system can achieve a court victory, what, precisely, is the minimum that the public is entitled to expect from a qualified human?

There is a temptation to treat such stories as novelties, amusing footnotes in the march of technology. That would be a mistake. The legal system is not just a market; it is an infrastructure of trust. If AI makes law cheaper and more widely available, it could alleviate a chronic national failure in access to justice. If it makes law more error-prone, more opaque, or more cavalier with confidential material, it could corrode confidence in institutions that already feel distant to many citizens. The Taskforce’s warning about striking off is, in its own way, an attempt to hold the line: to insist that innovation does not dilute duty.

The profession now faces a delicate redefinition of competence. In the short term, AI will be used unevenly, sometimes brilliantly, sometimes recklessly, often quietly. Some lawyers will adopt it with zeal and discover that speed is not the same as accuracy. Some will resist it and discover that resistance is not the same as prudence. Regulators, for their part, will be forced to make a series of judgements that are as much about culture as about technology: when is AI use responsible, when is it negligent not to use it, and what kinds of mistakes are no longer excusable in an era of machine assistance?

It is telling that the Taskforce has chosen to frame the issue in terms of liability and discipline rather than merely productivity. This is not a management seminar. It is a warning shot across a profession that prides itself on controlling risk. The message is that AI is becoming part of that risk landscape, both as a source of error and as a means of reducing it, and that the definition of “reasonable care and skill” will move accordingly. For lawyers, the safest position may prove to be neither blind faith nor stubborn refusal, but a rigorous, documented, sceptical competence with the tool that is reshaping their work.