Artificial Intelligence in Law
AI and the law
“A Valuable Assistant, but a Dangerous Substitute for Judgment…”
Artificial intelligence is rapidly becoming part of legal practice. It can review documents, summarise evidence, produce first drafts and identify possible lines of research in a fraction of the time traditionally required. For a profession facing rising costs, heavy workloads and a persistent access-to-justice problem, these capabilities are difficult to dismiss.
Yet law is not simply an information-processing exercise. Legal work requires accuracy, judgment, confidentiality and candour towards the court. Recent decisions in England and Wales demonstrate what happens when the speed and apparent authority of generative AI are mistaken for reliability. The emerging position is therefore neither that lawyers should reject AI nor that they should embrace it uncritically. AI has a legitimate and potentially transformative place in legal practice, but it must remain subject to meaningful human supervision and rigorous verification.
The case for AI in legal practice
The strongest argument for using AI is efficiency. Litigation and transactional work frequently involve reviewing large quantities of material, comparing clauses, constructing chronologies and producing routine correspondence. AI can assist with these tasks rapidly, allowing lawyers to devote more time to strategy, advocacy and advising clients.
This may also reduce costs. Legal services remain unaffordable for many individuals and small businesses, while overstretched courts increasingly encounter litigants who cannot obtain professional representation. Properly designed tools could help people understand procedures, organise evidence and prepare basic documents. They could also enable solicitors to handle lower-value matters economically rather than turning clients away.
The Solicitors Regulation Authority has recognised these opportunities. Its assessment of AI in the legal market identifies potential improvements in speed, capacity, affordability and access to services. It even suggests that firms may face a commercial risk from failing to adopt useful technology. In May 2025, the SRA authorised Garfield.Law, described as the first regulated law firm in England and Wales to provide legal services through a substantially AI-driven model. The service was intended to help small and medium-sized businesses recover unpaid debts through the small-claims process. Significantly, the regulator did not consider the use of AI itself inconsistent with professional practice. Its attention was directed instead towards safeguards, supervision and accountability. (SRA Risk Outlook; SRA announcement on Garfield.Law)
AI may also improve the consistency of routine work. A well-controlled system can apply an approved checklist to every file, identify missing information and compare documents against established standards. It does not become tired or overlook a page because it is working late. Used within secure systems and constrained by reliable source material, it can function as a second pair of eyes.
There is also a broader professional argument for adoption. Lawyers have always used technologies that alter how legal work is performed, from electronic databases to automated disclosure platforms. Refusing to use AI simply because it is new would protect neither clients nor the reputation of the profession. The proper question is not whether a lawyer has used AI, but whether the lawyer has used it competently.
The case against uncritical reliance
The difficulty is that generative AI produces plausible language rather than guaranteed truth. A system may present an invented case, inaccurate quotation or obsolete proposition with the same confidence and polished style as a correct answer. In legal practice, where a single authority can affect the outcome of a dispute, plausibility is no substitute for verification.
This risk was illustrated in Harber v Commissioners for HM Revenue and Customs [2023] UKFTT 1007 (TC). A litigant in person appealing a tax penalty supplied the First-tier Tribunal with nine supposed decisions that appeared to support her case. The decisions could not be found because they did not exist. The tribunal accepted that the appellant had not knowingly attempted to deceive it and considered that the authorities had probably been generated by an AI system. Nevertheless, the incident consumed judicial time and illustrated how readily fabricated law can enter legal proceedings. It also showed that users without legal training may be particularly vulnerable to convincing but false answers. (Case report)
The risks became still clearer in the Divisional Court’s decision in R (Ayinde) v London Borough of Haringey; Al-Haroun v Qatar National Bank QPSC [2025] EWHC 1383 (Admin). The court considered two separate proceedings in which false authorities had been placed before judges. In the Ayinde proceedings, written grounds prepared by counsel contained five fictitious cases. In Al-Haroun, a witness statement included numerous non-existent authorities and inaccurate propositions. The court dealt with the cases together because they raised a common and serious concern: legal material apparently produced or affected by generative AI had reached the court without adequate checking.
The judgment is important because it rejects the suggestion that AI changes the lawyer’s professional responsibility. A practitioner cannot excuse a false submission by explaining that it came from software, a client or a junior colleague. The court observed that publicly available generative AI tools are not capable of conducting reliable legal research and that their outputs must be checked against authoritative sources. It referred the lawyers involved to their professional regulators and warned that placing false material before a court may, depending on the circumstances, result in wasted-costs orders, disciplinary action, contempt proceedings or even criminal consequences. (Full judgment)
The central lesson of Ayinde is not merely that AI sometimes makes mistakes. Human lawyers also make mistakes. The particular danger is that AI can manufacture errors at speed, express them persuasively and conceal their artificial origin. A fabricated authority may resemble a genuine neutral citation; an invented quotation may sound judicial; and a false summary may combine enough accurate detail to escape superficial review.
Other concerns extend beyond hallucinated cases. Entering a client’s information into a public AI system may compromise confidentiality, legal professional privilege or data-protection obligations. The provider may retain prompts or process information outside the jurisdiction. AI-generated assessments may also reproduce biases found in training data. In criminal, immigration, family or employment matters, such biases may affect people who are already vulnerable. Meanwhile, overreliance on automated drafting may gradually weaken the ability of junior lawyers to research, analyse and write independently.
There is also a problem of transparency. A lawyer may be unable to explain why a system reached a particular conclusion or what information influenced it. That opacity sits uneasily with a profession whose decisions must often be justified to clients, opponents, regulators and judges.
Professional reputation and public confidence
The legal profession depends upon trust. Courts accept advocates’ statements about authorities partly because advocates are subject to demanding ethical duties. Clients disclose sensitive facts because they expect competence and confidentiality. The public accepts lawyers’ privileged role in the justice system because lawyers are expected to exercise independent judgment.
Unchecked AI use threatens each element of that relationship. A fabricated citation wastes the opponent’s time, increases the client’s costs and requires the court to investigate a problem that should never have arisen. More broadly, every publicised incident encourages the belief that lawyers are charging professional fees for unverified machine-generated work.
That reputational harm cannot be answered by blaming the technology. AI has no professional status and owes no duty to the court. Responsibility remains with the person who signs, files or relies upon the document. As the SRA has emphasised, firms remain accountable for AI outputs just as they remain accountable for work produced by their employees or contractors. The regulator’s approach to Garfield.Law is instructive: the system was not permitted autonomously to propose case law, clients retained control over procedural steps, and named solicitors remained ultimately responsible for the service.
A responsible place for AI
The appropriate response is controlled adoption. Firms should distinguish between low-risk assistance and work requiring intensive professional judgment. Using AI to format a chronology, suggest headings or summarise a document is not equivalent to relying on it for a definitive statement of law. The greater the potential harm, the stronger the required supervision should be.
At a minimum, responsible practice requires lawyers to:
verify every case, quotation and statutory reference against an authoritative legal source;
read the underlying authority rather than relying on an AI-generated summary;
confirm that the law is current and applicable in the relevant jurisdiction;
avoid entering confidential or privileged information into systems that have not been approved for that purpose;
review factual assertions against the evidence;
keep appropriate records of how significant outputs were produced and checked;
train and supervise staff in the limitations of the tools they use; and
ensure that a qualified person takes responsibility for the final work.
These precautions should not be treated as optional administrative burdens. They are the modern expression of long-established duties of competence, supervision, confidentiality and candour. Nor should AI be asked to certify its own work. Asking the same model whether a citation is genuine merely invites another plausible answer. Verification must involve an independent and authoritative source.
Conclusion
The debate about AI in law is sometimes presented as a choice between innovation and professional tradition. That is a false opposition. Used carefully, AI can make legal services faster, more affordable and more accessible. It can relieve lawyers of repetitive work and create more time for the human qualities that clients value: judgment, empathy, strategic insight and persuasive advocacy.
But Harber and Ayinde demonstrate that efficiency without verification can damage individual cases and the justice system itself. The danger lies less in the existence of AI than in the surrender of professional judgment to it. A lawyer may delegate a task to technology, but cannot delegate responsibility.
AI therefore has its place in law—as an assistant, a starting point and, in appropriate settings, a powerful means of improving access to justice. It must not become an unquestioned authority. The future reputation of the legal profession will depend not on whether lawyers use AI, but on whether they remain recognisably professional when they do.
