The spread of AI systems appears increasingly central to the evolution of work and professions, and lawyers are not exempt. This evolution directly calls on law schools on a global scale to reflect on legal education. Even if the most visible institutional responses to generative AI in legal education have so far come from US law schools, such as Berkeley,¹⁾ Columbia,²⁾ University of Chicago,³⁾ the wider picture is increasingly more varied. IE Law School in Madrid has an initiative to embed AI across every stage of its programmes;⁴⁾ Catolica Global School of Law and the National University of Singapore have been working with new partnerships with legal tech providers;⁵⁾ Indian law schools have established dedicated research centres on legal tech.⁶⁾ These approaches are pilots rather than prohibitions, and they show that the question is being taken seriously in very different systems.

However, what remains rare, almost everywhere, is a broader strategy taken at the institutional level about the mission of academic institutions in the age of AI. Particularly, despite some exceptions, universities have acknowledged the relevance of AI, but legal education, particularly in Europe, is largely public, framed by national curricular requirements, and followed by a separate stage of professional training through bar schools and traineeships. Changes of this kind require universities and law schools to coordinate with ministries and bars, and this takes time that the technology has not allowed.

Nonetheless, students are not waiting, and neither is the legal market they will enter. The question raised by the role of AI in shaping legal practice is indeed institutional, and it addresses three dimensions of it, namely what students now expect, how the curriculum and its partnerships should be redesigned, and what the legal market is actually asking law schools to produce. The underlying idea is that law schools should neither prohibit nor simply adopt, but decide, and that this decision belongs to institutions rather than to the preferences of a single law school or even to the activities of individual instructors.

Students Skills and Expectations

The diffusion of AI through society has been faster than any institutional academic response, and universities encounter it first through their students. As an example, the Thomson Reuters Institute’s 2026 Law Student Pulse Survey underlines students’ awareness about the relevance of AI for lawyers.⁷⁾ Three-quarters consider learning to use AI essential for their profession. At the same time, almost the same proportion believe that overreliance will prevent students from developing core legal skills. Both propositions underline the main challenge and expectation for legal education, and students appear to have arrived at it without training.

Roughly a third say their school does not give them the skills their careers will require. Almost half report that policies vary across different courses based on instructors’ decisions, leaving no coherent institutional signal about what responsible use looks like, and this is often connected to the lack of guidance by academic institutions at the central level. The problem is not that law schools have chosen the wrong policy, but that, in most cases, universities first, and then law schools, have not decided, and have delegated to a single academic, and particularly course instructors, a decision that contributes to shaping legal education.

This has consequences for the attractiveness of law schools that go beyond recruitment marketing. A law degree presented as a sum of courses, some also on AI, and exams based on producing texts that a model generates instantly is increasingly a weak proposition, and prospective students are perfectly capable of noticing. Yet the answer is not to compete by advertising AI courses. Expecting AI to be part of legal education does not seem to be negotiable, but matching market expectations tends to set aside a mission that includes forming expectations, not merely satisfying them.

Nonetheless, the realistic position lies between the two positions. Law schools should be explicit about their approach to legal education, which requires explaining why certain competences are protected from automation during training precisely because they will be needed to supervise it afterwards, and this explanation has to be institutional rather than left to the willingness of each professor.

Redesigning the Curriculum, and the Partnership Question

Revising the curriculum means more than adding an elective on AI ethics or adopting a policy on students’ assessments. This could be a pilot, and some universities are already moving in this direction, particularly by building collaboration with providers of legal tech such as Harvey and Legora. However, the approach should expand to the whole curriculum and, at least, some elements deserve to be distributed across the degree rather than confined to a specialised course.

The first is a technical minimum. Students do not need to build models, but they need to understand why a system that predicts text produces confident mistakes, or why the same prompt yields different answers. Some schools have gone further than most law faculties would consider necessary, as Bucerius in Hamburg teaches introductory computer science and data science to its undergraduate law students on the view that computational thinking is part of the equipment.⁸⁾ The second element is advancing critical skills in verification and supervision. Much of what junior lawyers will do is read work they did not write and decide whether to stand behind it, and that competence can be assessed directly by giving students an AI-drafted memorandum containing a plausible but misdescribed authority and grading them on whether they catch it. The third is the professional and deontological frame. In France, the CCBE’s guide on generative AI locates these tools within existing duties of confidentiality and independence,⁹⁾ and Singapore’s Ministry of Law has issued a sector-wide guide that treats verification and oversight as professional obligations rather than good practice.¹⁰⁾ Students should learn early that signing an AI submission is an assumption of responsibility that cannot be delegated to a machine.

Case law has begun to say the same thing, and to say it in terms that concern legal education directly. In Ayinde,¹¹⁾ the Divisional Court considered two matters in which fabricated authorities had been placed before the court, and referred the lawyers involved to their regulators. This case, together with others, shows the need for training which cannot be limited to courses about AI but requires a broader reframing of legal education in the age of AI.

Integration on this scale requires infrastructure that faculties do not possess, which explains the rapid growth of partnerships with legal technology providers. Harvey’s academic programme began with six US law schools and has now expanded on a global scale.¹²⁾ These arrangements solve a real problem, since students cannot learn to supervise tools they never touch, and this can provide students a competitive advantage. But they also raise a governance question. When curricular content is co-developed with the supplier of the platform, and when the free licence expires after graduation, private actors acquire influence over professional formation that attaches little to public educational institutions.

However, not very different from using Zoom or Teams for providing distance learning in these years, alternative solutions coming from academic institutions are scarce or mostly not competitive. However, they are not absent. A group at Western Sydney is building its own teaching tools rather than licensing them,¹³⁾ including one designed to train students to spot hallucinated material, which is precisely the competence the case law now demands. The realistic response is therefore not refusal but conditions, which means partnerships should be plural rather than exclusive and premised on the faculty to keep oversight on what is taught.

Facing the Market, and the New Roles

The legal market is sending two signals at once. A general concern about the role of AI in replacing trainees, often depending on the scale of the law firm or the institution, is spreading. Historically, junior lawyers acquired training and judgment mostly through repetitive actions.¹⁴⁾ If the profession no longer trains through repetition, the burden and competence lock upstream. The university may become the last institution where anyone is required to read an entire judgment slowly and to critically examine legal concepts, and law schools should treat that as their comparative advantage rather than as an anachronism to be modernised away.

At the same time, the market is creating roles that did not exist when most curricula were designed. Legal engineers, legal architects, AI governance counsel, product counsel in legal technology firms and compliance specialists for the AI Act all combine legal reasoning with an understanding of how systems are built and procured, and none of them requires the ability to write code. The sector itself has become a serious destination, with legal tech companies such as Legora building substantial businesses, and law graduates increasingly appearing among their founders rather than only among their customers.

Nonetheless, legal education has traditionally prepared students for the bar, the bench, the ministry and the firm, and it is poorly equipped to prepare founders. Clinics, legal design laboratories and access-to-justice projects are the natural vehicles for this, since they teach students to build something that must actually work for a user, but they are often the exception. Part of the explanation is institutional rather than intellectual. Teaching innovation of this kind requires a strategy and resources to invest, while often it does not count in research assessment, thus depending on individual academics willing to absorb the cost.

This evolution requires law schools to reflect more on how they can move from experiments to strategy in legal education. Pilots are a first step, but academic institutions are required to address the questions raised by AI, focusing on students’ expectations and market demands. What is missing is not the decision to use them, but the recognition that declining to decide or just focusing on experiments is itself a choice.