The Duty to Use - Artificial Intelligence and the Reasonable Lawyer

​​

In July 2026, the UK Jurisdiction Taskforce warned that a professional may be liable not only for misusing artificial intelligence but for failing to use it where a reasonable practitioner would have. The warning, though cautiously worded, marks a genuine inflection point. This article frames AI not as a substitute for legal judgment, but as a tool that magnifies legal knowledge, assists legal drafting, capture knowledge and then help lawyers refine the craft of lawyering. Learning law has always been a lifelong exercise because legislation, courts and political change keep remaking the field. AI compresses that cycle. Read alongside emerging judicial guidance in Singapore and India, the direction of travel is unmistakable. The article argues that the professional duty of legal competence is a moving standard, and it is moving towards acceptance of AI in the practice of law.

I. A Warning Worth Reading Twice

Every profession has its quiet revolutions, and they rarely announce themselves in the language one expects. The latest arrived not as a manifesto for machines but as a careful sentence in a technical paper. In July 2026 the UK Jurisdiction Task force (UKJT) observed that “a professional could also be liable for failing to use AI for a task when a professional exercising reasonable care and skill would have done so.”

The line was picked up widely since the UKJT is a body backed by the Ministry of Justice in UK and chaired by Sir Geoffrey Vos, the Master of the Rolls, and The Telegraph framed it as lawyers risking being struck off. The specialist pressstatedit soin near-identical terms.

Readcarefully,it is double-edged.

A lawyer may be sanctioned for misusing AI as the recent run of pleadings marred by fabricated citations and fictitious authorities is warning enough.But a lawyer may equally, the Taskforce suggests, be sanctioned for declining to use it. Matthew Lavy KC, who chaired the drafting committee, put the point with disarming economy: “We are not far from a time where harm caused by a failure to use AI where a reasonable lawyer would have done so will also be actionable.”

That is why the better frame is not “AI versus the lawyer.” AI is better understood as an accelerant to the ‘craft of lawyering’ and to the ‘democratization of access to law’. To be clear, most lawyers of my generation have never practised in a static discipline. They have had to keep up with changing law and relearn the application of statutes such as the Telecom Act, the Companies Act, 2013, data protection laws and the new criminal codes.

A serious practitioner has always had to relearn the law as legislatures amend it, courts reinterpret it and governments of different ideological temperaments redirect its priorities. What once shifted over three or four decades may now shift within a single term of a party in office. Recent developments in both the United States and India show how quickly legal priorities can shift when elected governments use legislative majorities and executive power to reshape policy around a governing ideology. Governments with legislative majorities often reshape law around their policy and ideological priorities, with consequences that apply across the population. AI does not create this instability; it compresses it, understands the new legislations and then offers the profession a way to keep up.

II. An Old Standard, Newly Applied

It is tempting to read the guidance as an endorsement of legal artificial intelligence. It is nothing of the sort, and its restraint is precisely what makes it formidable. The UKJT invented no new liability regime. Its Legal Statement on Liability for AI Harms concludes that existing English law already supplies a coherent framework (as its earlier statements on crypto assets and smart contracts did), statements later adopted directly by English judges. No new legislation, the Taskforce says, is required.

What the guidance does is apply a very old idea to a very new instrument. Whether a lawyer has breached a duty by failing to use AI is to be judged, in the Taskforce’s words, by whether “a reasonable professional of a comparable rank/specialism should have used AI in that context.” AI, in a practitioner’s hands, is a tool; and the question of whether a tool should be used is no different in principle from the question that arises for any other tool at the professional’s disposal. The Taskforce offers a concrete illustration: a solicitor in the Business and Property Courts who fails to advise a client to consider AI-assisted review of a large volume of documents may already be exposed. Reasonableness, in other words, is a moving target and it is the profession’s own competence that moves it.

Seen this way, the duty to use AI is not a fashionable technology claim. It is an extension of the older duty to remain competent in a changing legal order. If the law itself now changes faster, and if the sources of legal meaning multiply faster, then the reasonable lawyer’s toolkit must also change. Competence is no longer measured only by memory,seniorityor access to relationships; it is increasingly measured by the ability to use new instruments to find, test and improve legal judgment.

III.  The Deliberate Practitioner

If AI is a force multiplier, and in capable hands it plainly is, then the choice of which tool to multiply with can no longer be made once and quietly forgotten. The marketplace is unsettled, the capabilities shift from one month to the next, and the honest consequence is an uncomfortable one: the lawyer must now make deliberate, repeated decisions about the instruments in front of him, not once, at the level of the firm, but matter by matter, and on the harder files task by task. Few practitioners are yet engaged at that level of attention. The better ones will get there, and for the oldest of reasons, because competition will demand it, and because there is every prospect the profession will one day require it of them.

The shape of that requirement is already visible across the Atlantic. The State Bar of California, in its 2026 practical guidance on the use of generative AI, frames the obligation not as a single duty but as two:

“When using AI, a lawyer has two related duties of competence. First, a lawyer must attain baseline technological competence by developing a reasonable understanding of the relevant AI system’s capabilities, data sources, limitations, and material risks and benefits before deploying it in connection with legal services. Second, a lawyer must exercise independent professional judgment by reviewing, verifying, and correcting AI-generated outputs consistent with the learning and skill reasonably necessary for the representation. Competent use of AI-enabled tools, therefore, requires not only familiarity with the technology itself but also the ongoing application of the lawyer’s independent professional judgment to all outputs used in providing legal services.”

— State Bar of California, Practical Guidance on the Use of Generative AI in the Practice of Law (2026)

Read that standard closely and it sets a genuinely exacting test: a “reasonable understanding of the relevant AI system’s capabilities, data sources, limitations, and material risks.” It is fair to ask whether we can honestly reach it while taking a vendor at his word that his particular wrapper is the finest on the market. The underlying models are already black boxes. Are we now to accept black-box wrappers bolted on top of black-box models, and dignify the arrangement with the name of diligence? The question is not rhetorical, and I do not pretend the profession has yet answered it.

IV. The Strongest Case for Caution

The sceptics deserve a fair hearing, and at their best they have one. The objections are serious and should not be waved away.

Reliability. Generative models still fabricate. The catalogue of lawyers filing hallucinated citations is real, embarrassing and, in places, sanctionable. A tool that invents authority is a tool that can end a career.

Provenance and confidentiality. The lawful basis of some training data remains contested, and the temptation to feed privileged material into an insecure system is a professional hazard of the first order. The UKJT itself warns that putting a client’s confidential or privileged information into an AI system that is not suitably secure is highly likely to be a breach of duty.

Concentration and cost. Vendor lock-in, opaque pricing and the geopolitics of a handful of dominant model providers are not trivial concerns for a profession that prizes independence.

Deskilling and access to justice. There is a genuine worry that the junior who asks the model before thinking will never learn to think, and a further worry that automated decision-making could entrench, rather than relieve, existing inequalities in the system.

These are not the complaints of Luddites. Indeed, the same UKJT guidance that contemplates liability for non-use is unsparing about misuse: it warns of negligence in failing to conduct due diligence on a new or untested system, in failing to explain to a client (at least in broad terms) how the AI works, and in failing to exercise oversight of its outputs for error and bias. The duty cuts both ways, and rightly so.

The anxiety is therefore not merely technical. It is about professional status, access and control. It is about preserving the status-quo.

A profession that once derived value from scarcity of access (to libraries, seniors, precedents, clerks, database sand networks) is now confronting a technology that makes large parts of legal knowledge and first-draft creation widely available. Today, clients can test a lawyer’s understanding of the law in much the same way patients now arrive at medical consultations after reading AI-generated explanations of their symptoms.

That transition is uncomfortable precisely because it changes where lawyers capture value. The question is whether the profession responds by defending old scarcity, or bymoving up the chain into judgment, strategy, ethics and craft.

V. Every Objection Is About How AI should be used

Set each of the above objections side by side and a pattern emerges: every one of them is an argument about how AI should be used. Taken seriously, they point toward disciplined use, not professional abstention.

Hallucination is a reason to verify, not to abstain and the profession already strikes people off for filing fictitious citations, whether a junior invented them or a model did. Provenance and vendor risk are procurement problems, solved by contract, diligence and closed, dedicated systems rather than consumer chatbots. Confidentiality is an information-security discipline the profession has practised since the first locked filing cabinet. Deskilling is a training-culture problem: a firm that bakes verification and independent judgment into its habits produces better lawyers, not worse ones. Law firms which ensure that they only use legal AI tools grounded on knowledge graphs and semantics are taking better precautions than using plain generative AI models.

None of this is unprecedented. The profession has made versions of this argument before: that electronic legal databases such as Westlaw, Lexis and Indlaw would weaken memory; that electronic disclosure would overwhelm juniors; and that online filing and email would compromise accuracy and discipline. Each anxiety contained a real professional concern. Each tool eventually became part of competent practice. The pains of adapting to a new tool are not an argument against the tool; they are a matter of change management.

For the lawyer, AI is already beginning to operate in three phases.

First, it organises knowledge: Statutes and regulations are found in databases like LawBase. Judgments, cases, commentaries and pleadings are found in tools like CaseBase. Contracts, drafting tools and institutional memory are found in tools like Dealbase.

Second, it assists creation: Today first pass drafts, chronologies, issue lists, cross-examination notes, research memos and transaction documents can be created on the fly over the various LexOps tools.

Third, and most importantly, it improves the craft of lawyering by helping lawyers test alternatives, expose weak assumptions, compare lines of reasoning and refine work that would otherwise remain trapped inside the limits of one mind or one team. The LexOps and Diginote tools are all efficiency tools which could help lawyers improve on specific elements of their practices.

The first two functions were always available, at least to those with money, pupillage, institutional support and access. The third is where the profession may be remade, because it shortens the path to the kind of comparison, self-correction and refinement that once came only through years of practice and access to competent teams which only top tier 1 firms, inns or magic circle law firms could afford.

The Radiologist example– The guidance itself supplies the decisive image, and it is worth dwelling on. A lawyer who rejects AI, the panel suggested, should be viewed much as we would view a radiologist who refused to use tools now extremely effective at identifying cancerous tumours.

Consider what that analogy demands of us. If a radiologist declined software proven far better than the unaided human eye at catching a tumour, and a patient’s cancer went unread as a result, we would not call that principled caution. We would call it negligence, and we would be right. Legal practice is beginning to move in a similar direction.

When a model can surface the one decisive authority or piece of evidence buried on page twenty-four thousand of the disclosure, or the single inconsistent line in a witness statement that unlocks a cross-examination, the lawyer who refuses even to look is not protecting the client. He may be failing that client, and billing for the additional hours his refusal required.

VI. Times Are Changing 

The English position is not an outlier but a bellwether. Look east, and the same instinct is at work in jurisdictions no one would accuse of recklessness.

Singapore. Since 1 October 2024, the Guide on the Use of Generative Artificial Intelligence Tools by Court Users has governed proceedings across the Supreme Court, the State Courts and the Family Justice Courts. Its stance is studiously neutral: it does not ban the technology. It permits it by extending even to the production of first drafts of witness statements on the firm condition that court users take full responsibility for the accuracy and appropriateness of anything they place before the court, and it declines to require litigants to declare when such tools have been used. The obligation is not abstention; it is ownership.

India. In June 2026the Supreme Court of India released, for public consultation, its draft Regulations for Use of Artificial Intelligence in Courts, 2026. The draft is emphatic about limits. It insists on human primacy and prohibits AI adjudication outright. But it does not prohibit the use of AI for legal work. On the contrary, it expressly contemplates AI-assisted research, drafting assistance, citation verification, translation, transcription and case management, subject to disclosure and oversight. The most careful judiciary in the common-law world’s largest democracy is not asking whether AI belongs in legal research. It is asking how to govern it there.

That is the tell. Across three very different legal cultures, the regulatory question has shifted in the space of a few short years from “should we permit this at all?” to “how do we channel it responsibly?” When the framing changes that decisively, the underlying answer has usually already been given. 

That shift also explains why the debate feels larger than legal technology. Courts and regulators are beginning to accept that the newer legal approaches, workflows and solutions to address judicial delays will be built on AI-assisted research, verification and drafting. The law’s breakthroughs have always depended on the tools available to discover patterns, retrieve memory and test argument. AI shortens the time between legal research, drafting and strategic judgment. It brings tomorrow’s research cycle closer to today’s file especially in legal workflows. 

VII. From Optional Tool to Professional Expectation

Let me be precise about what I am and am not claiming. The day of universal compulsion has not yet arrived; a lawyer today is not negligent merely for preferring the old ways on a given file. But it is coming, and sooner than the comfortable suppose. The duty of competence has never been frozen in time. It tracks, as it always has, what a capable practitioner can do for the people who trust them with their affairs and, sometimes, their liberty. 

Today, AI tools reliably deliver better, faster and cheaper outcomes for clients especially in the grinding document-work of contracting or modern litigation. As AI tools demonstrably function effectively in some cases, the refusal to use them stops being a matter of personal taste and becomes a breach of duty. Tomorrow’s professional misconduct will not only be the lawyer who lets an AI hallucinate a case into a pleading. It will equally be the lawyer who, out of pride or inertia, left the better tool in the drawer and let the client pay the difference.

The deeper economic transition is already visible. What once took years of apprenticeship, access and connections to acquire, the ability to locate the right authority, assemble the first draft and compare competing legal positions is being democratised. That does not make lawyers less valuable. It changes the basis of their value. The lawyer of the next decade will be judged less by possession of information and more by the quality of supervision, synthesis, strategy and judgment brought to information that many can now reach.

Verify everything. Trust nothing blindly. But use the tools that competent practice requires. AI is not the end of lawyering; it is one more reason lawyering will become more demanding. As legal knowledge becomes easier to access and first drafts become easier to produce, professional value will move toward supervision, synthesis, strategy and judgment. The lawyer who refuses that shift will not look principled for long. He will look unprepared.