Justice Aidan Xu: Opening keynote address at the 7th Lawtech Forum Asia

Opening Keynote 

The New Legal Order: AI, Innovation, and the Road Ahead for Asia

Thursday, 21 May 2026

Justice Aidan Xu

Judge of the General Division of the High Court,

Supreme Court of Singapore

I. Introduction

1. This year, we in Singapore celebrate two hundred years since the founding of our courts through the Second Charter of Justice of 1826. Within that span, we have moved from a rudimentary, makeshift system, to a colonial structure, and then to a vibrant, independent centre of common law doctrine and adjudication. We in Singapore, and indeed across Asia, have in that time faced and overcome many challenges. That puts us in good stead, I believe, to confront the challenges now posed by AI to our legal frameworks and systems.

2. What those challenges look like is, by now, well known. Just in 2022, four years ago, if a legal associate were asked to research a memorandum to advise a client, she would have logged onto a legal research site, considered her search terms, worked out the right Boolean operators, perhaps used some natural language processing, all to work through a single issue: what counts as consideration, for example. She would then have had to assemble everything, organise it, and address the client’s needs. She would have put herself into the shoes of the client, and perhaps more importantly into the shoes of the partner reviewing her work. She would then have typed everything out, proofreading for grammar and spelling, and for compliance with firm style. The partner would have then reviewed and critiqued the memorandum, rearranging paragraphs, identifying errors, recalling perhaps some other authority that was missed out, or some relevant statutory provision. Then it would have been emailed out to the client, who would have read it, or put it on file. At the end of the month, the associate would have tallied up her billable hours, and the firm would have billed the client.

3. In 2026, the likelihood is that a lot of that work has been cut short. The associate, if she has access to one of the legal AIs, such as Legora, CoCounsel or Harvey, or to a paid version of Claude or Perplexity or ChatGPT, could have typed in the exact scenario presented by the client, and had a fully formed answer generated in a matter of minutes, if not seconds. She may or may not verify what has been produced. She would not have had to exercise her mind about organising things, weighing what to put in or leave out, or thinking about what the client wants. The partner may, or may not, do a cite check. The client would receive the memorandum, see the em-dashes abounding, and decide perhaps, that he is better served by using AI himself, rather than paying an outside lawyer to do just that. 

4. That is one possible reality in 2026. And it illustrates the changes that have been wrought in just four years. The expansion and growth in AI capabilities continues: just over the past few months we have had a massive burst in agentic capabilities as well as deployment at least in some countries, by ordinary citizens on a surprisingly large scale. New frontier models such as Mythos¹ and its ilk may pose significant challenges for many of our institutions. As it is, we are still grappling with these challenges within our own cities and countries. We have barely begun to think about what they mean for the legal world in Asia or the world.

5. There are indeed many challenges facing the legal world as it engages with AI. I would like to propose that there are three convergent forces at play. First, an increasingly capable and powerful AI, whether through sheer compute or through the use of agents. Secondly, the changing expectations of the users of legal systems, that is clients, parties, and citizens. Thirdly, the increased porousness of our national borders.

6. With that in mind, let me set out what I hope to achieve this morning. I will first try to sketch the trajectory of AI in legal work, and where the line between machine output and human contribution still lies. I shall then turn to the likely consequences for the legal profession, including pressure on work productivity and on the development of future lawyers. I will address the position of those who are lawyer-less, whether by choice or by circumstance and what this means for our courts. I then hope to draw out what I see as the Asian dimension. These Asian borders may be more porous, but our markets remain uneven. And I shall then close with what I think is required to navigate this terrain, and what it means for all of you involved in legal innovation.

II. AI in Legal Work

7. Let me begin with where things are. There are, and no doubt will continue to be, many horror stories about AI in the law. I was just last month, at an insolvency conference in London, where the big news going around was the unfortunate use of hallucinatory AI output in submissions to a bankruptcy court.² We have had similar instances of hallucinations and other inaccuracies in our courts here in Singapore. So has India, Canada, and the United Kingdom, and really almost everywhere else.³ The point is perhaps an easy one to make, that AI in legal work is unreliable and cannot be trusted to tell a real case from an imagined one.

8. But we in this room know that this is not the complete picture. As prefaced in my illustration above, much legal work can be done, accurately and quickly, on AI: legal AI such as CoCounsel, Legora, Harvey and others can work very well, with curated databases. For routine tasks, they often produce work no worse than a competent junior associate would, and often very much better. Indeed, even the major general-purpose models, such as Claude, ChatGPT, Gemini, DeepSeek and others, have been getting steadily and substantially better, especially when given good prompts and asked to link to source materials for verification. The horror stories have not disappeared, and mistakes will still arise, but I would say not particularly any worse than what could be expected when lawyers are rushed, or do not stop to check.

9. That improvement has particular consequences for our region. Cross-border law in Asia, which until very recently was the preserve of specialists with library, language, and contact privileges, is now more readily researched. A practitioner in Singapore can, in minutes, get a working understanding of a point of Vietnamese contract law, or Indonesian competition rules, or Korean data protection regulation. The work may not be perfect. But it is good enough to start a conversation with the client, to brief a team, and to identify the right local counsel to engage. Regulators across our region may try to limit and control this, but the reality on the ground is that the practice is already happening at scale. There are blind spots, of course, where training data is thin or local procedural rules are particular, but these gaps too are being addressed year on year. The same applies to language. Translation capabilities for Indonesian, Vietnamese, Thai, and Tagalog are now at a level that would have seemed implausible five years ago. The baseline of accessible regional knowledge is rising rapidly, and the consequences for how we practise law across borders are profound.

10. As again everyone here would know, reliability and capacity are both increasing. We see marked improvement in accuracy and decline in hallucinations and inaccuracy. Agentic capacity has also changed the game. Systems that two years ago would respond once to a prompt and stop are now capable of planning a multi-step task, executing it, verifying their own work, and reporting back. The combined effect is that the question, for a meaningful set of legal tasks, has shifted. It is no longer whether AI can perform them. It is who still does them manually, and why.

11. There remain some areas which may yet fall outside. It may be that legal reasoning from precedent and principle will continue to be the province of human lawyers and judges. I am not sure though that there is such a bright line between explanation, which AI is supposedly not good at, and the semblance of explanation, which I think they can do, and continue to improve at. It may also be a while before machine judging is considered acceptable. Gen X may be strongly against, Millennials may be queasy, and Gen Z may be resentful, but we should not think ourselves permanently protected from greater displacement. Gen Alpha and Beta may yet experience a vastly different world and have different viewpoints.

III. Displacement, Pressure, and the Formation of Expertise

12. That lets me turn next to the other aspects of the new legal order: displacement, pressure, and the impact on the institutional fabric of our profession. Dire forecasts for the impact of AI on professional, white-collar work have been made⁴. The estimates vary, but the direction is clear, and the time horizons are short.

13. We need to respond to this challenge and prepare. We do have a challenge in training and work within the legal profession, whether in-house or at the bar. AI is widely promised and sold as an instrument of efficiency. In the experience of many firms, it is instead delivering pressure and stress. AI does not simply automate routine work and free firms to redeploy capacity. Just as the washing machine and vacuum cleaner did not give free time to the housekeeper, AI gives efficiency with one hand but takes away with the other. AI accelerates the cycle in which competitors must deploy. It creates real and rising costs in tooling, training, security, and governance. There is at least a perception that AI has not, for most firms, been the cost-saver that was promised. It is, for many, an expense centre that is also a competitive necessity.

14. The pressure is also asymmetric and jagged. It falls differently on different parts of our profession, on different practice areas, and on different generations within a firm. The global, multi-national firms with deep resources can absorb the cost of enterprise-grade systems, they have the luxury of experimentation, and can with time harness AI properly. The truly local one person operation can perhaps hold its position, for now, on close client relationships and on a lean cost base. Volume work, that is conveyancing, due diligence, document review, and basic research, is most directly automated. Bespoke advisory work, and complex contentious work, less so. The senior lawyer who can supervise AI output is in a different position from the junior lawyer whose first years of practice were once spent doing the very work that AI now does best. 

15. In my capacity as Chair of the Promotion of Legal Technology Innovation Committee of the Singapore Academy of Law, I am working with my colleagues on gathering information about the impact on law firms and young lawyers, but my sense is that mid-sized firms feel it most acutely. They lack the scale of the largest firms, who have the resources, but they also lack the lean cost base and different expectations of small local practices. They are caught in the middle. 

16. The pressure on the profession does not stop at the firm or organizational level. The Chief Justice flagged at the Opening of the Legal Year 2026 that some sixty per cent of our young lawyers, surveyed at recent mass calls, indicated they were likely to leave private practice within five years.⁵ The various possible reasons such as workload, work-life balance, and challenges at work, are not new. But they are intensifying, and AI is part of the reason. The new committee that the Chief Justice and the Minister for Law are co-leading an examination of attrition and the impact of AI on practice. That is one institutional response. 

17. The third area of challenge that we need to deal with is how expertise will be formed in this new environment. This is a wicked, complex and tangled problem. Training through routine work has been the process through which expertise is honed in the legal profession, as it has been in other professional settings. Junior lawyers acquired skill and judgment by doing the things that now AI does better. As AI spreads, the path to becoming an expert or indeed a competent lawyer becomes less obvious, and harder to sustain. 

18. One possibility is to introduce some form of friction requiring tasks being done manually, much as we had to learn much arithmetic before being allowed to use calculators at school. That would be good. But I fear that it would not be practicable. It would take a very confident, very secure senior lawyer, faced with increased pressures from clients and from competition from others, to hold off and keep her young lawyers away from AI use. I do suspect the dam will break, but I am afraid I have no ready solution. Ideally, training and formation should be done in a longer, deeper stint in a controlled setting, just focused on practice skills. That would however prolong the formation. I do not right now know if we can find a happy medium. I understand the Academy of Law is exploring solutions. All I can posit is that we need to continue to rethink deeply what professional formation entails, and we may need to face up to very substantial structural changes. 

19. That kind of challenge will be faced across this region. While the apprenticeship model has different variations even in the common law systems, and significant differences when we look at civilian and socialist systems, the pressures are common. As in other areas, sharing of lessons will be useful and should be encouraged.

20. Another point of pressure is that AI is changing client expectations of what lawyers do. The knowledge gap between lawyer and client is decreasing. Clients arriving at the lawyer’s doorstep today are substantially better informed than they were even three years ago. They have used the same tools we have. They have increasingly obtained AI advice before they walk in. The lawyer becomes an AI fact checker.

21. What can the lawyer do in this environment. I believe there is some value in lawyers seeing themselves not just as lawyer-technicians, meaning giving advice just on the law or evidence. There is still, at present, room for growth in acting as counsellors: giving broader advice, and adding to the human component, in terms of addressing underlying issues, encouraging conciliation, or non-technical solutions. The only question is, I do not know how long this will remain the exclusive purview of human lawyers.

22. This then brings me to one of the implications of changed client expectations. The rise of the lawyer-less.

IV. The Rise of the Lawyerless

23. I come now to a development that is, in my view, among the most consequential of all that I have described. The availability of generative AI could lead to an increase in the lawyer-less, or at least the lawyer-light. The phenomenon takes two forms. There are the self-represented persons in litigation. And there are the corporate clients who, on the transactional side, are turning to AI before they turn to lawyers. Both deserve our attention.

24. Taking first the self-represented person in litigation. It is likely that numbers will continue to rise. AI is one visible driving force, encouraging more litigants to pursue proceedings on their own, though increased costs, as well as more legal regimes excluding lawyers, would also contribute. Let me focus on AI for the purposes of this speech. AI has potentially equipped these litigants with capabilities they never had before. They can draft pleadings. They can frame legal arguments. They can engage with complex procedural rules. The result, across our region, is that we are seeing more self-represented persons, in more types of case, with more sophisticated submissions, than ever before. Of course, mishaps can and have occurred. But, as I have noted above, with the increased capabilities of AI, it is likely that such mishaps and misuse will decrease.

25. Our effort to harness AI for good for the self-represented is evident in our work in the Small Claims Tribunals, where we have partnered with Harvey AI on a pro bono basis to explore how generative AI can assist self-represented litigants.⁶ We chose this venue deliberately, because the Small Claims Tribunals are where the justice gap is often widest. These are cases that are factually and legally simple, but where the disproportionate cost of legal services places assistance out of reach. Our initial deployment has been measured. We began with on-demand translation of court documents into Chinese, Malay, and Tamil. In September last year, we announced that we would extend the use of generative AI to summarise the parties’ documents, so that litigants could better understand each other’s positions, and, we hoped, reach an amicable resolution.

26. What I have noticed, and what others in this room may have noticed too, is that the lawyerless tendency is not confined to litigation. We are seeing it, perhaps more quietly but no less powerfully, on the transactional side. Aside from self-represented persons, the anecdotal evidence is mounting of an increased readiness on the part of corporate clients to have less human lawyering. Businesses turn to AI first, to draft a contract, to review a clause, to summarise a regulatory position. They then come to lawyers to check AI’s work, rather than to do the original analysis. And they expect to pay less for such checking than they once paid for the underlying work. The pattern is uneven across sectors and across jurisdictions, but it is real, and it is gathering pace.

27. That raises a question of some importance for those of us who care about the profession. What, then, is to be the scope of legal work going forward? My own answer, which I hold with some humility, is that lawyers will increasingly find their value not in the technical execution of legal tasks, where AI is rapidly closing the gap, but in the more distinctively human dimensions of advice. They will need to be counsellors. They will need to advise in a multi-disciplinary way, on both legal and commercial matters, on risk and on relationships, on what is technically permissible and what is institutionally wise. They will need to take responsibility, in a way that AI cannot, for a recommendation given to a client who needs an answer. The work that remains distinctively for the human lawyer is the work that requires judgment and discretion.

28. There is, however, a more constructive side to this story, and it deserves equal attention. AI’s impact on technical legal work may also open up areas of business that simply did not exist before. Legal services that were once uneconomic to provide, because the cost of human lawyering was too high relative to the value of the matter, can now be delivered at scale. We saw the early shape of this last year in the United Kingdom, when the Solicitors Regulation Authority authorised Garfield Law as the first AI-driven law firm in England and Wales.⁷ Garfield specialises in small-claims debt recovery for small and medium-sized enterprises, with claim sizes up to ten thousand pounds. Its pricing is striking: as little as a few pounds for a chaser letter, and modest amounts for the steps that follow. The firm is not autonomous, in the sense that each step requires the client’s approval, and named regulated solicitors remain responsible for its outputs. But it is a law firm whose central engine is artificial intelligence rather than human associates. We are seeing similar ventures in our region, including platforms such as Flash Law and others. The model is broader than any one company. It is the model of legal services delivered, in much the same way as software is now delivered, as a service. Law as a service.

29. The implications, for access to justice and for the shape of our profession, are considerable. Whether these particular ventures prove durable in their current form remains to be seen. The Asian context will, in any case, present its own obstacles to scaling, and I shall come to those in a moment. But I want to leave you with this thought before I move on. The architecture of a different kind of legal market is being put in place, in real time, in front of us. It will be uneven. It will be contested. Some of it will fail. But it is, I believe, the genuine harbinger of a future in which legal services are delivered very differently from the way most of us were trained to deliver them.

30. What I would add, by way of regional observation, is this. The squeeze on external counsel that corporate counsel are increasingly empowered to apply is the other face of the cost-and-time pressure I described earlier. From the firm’s side, this means insourcing of routine work, fixed-fee discipline, and AI-driven price transparency. From the corporate counsel’s side, it means greater scope, sharper expectations, and a heavier governance load. The two sides will need to renegotiate the division of labour with greater openness than has been customary.

31. In a number of our regional markets, the historical role of mid-sized regional firms is exposed to this squeeze. These firms have been the institutional vehicle through which Asian commercial practice has matured over the last two decades. The shape of that ecosystem five years from now is, I believe, an open question. That brings us to the specific challenges that we find in Asia. 

V. Borders That Blur, Markets That Differ: The Asian Dimension

32. The simultaneous blurring of our borders and the continuing divergence of our markets is, in my view, the most distinctively Asian dimension of the new legal order. We do not have the unified language of United States or North American market, nor the common legal framework and shared history of Europe. Our legal map is, one would dare say, fragmentary. Asia is home to common-law jurisdictions, that is Singapore, Hong Kong, Malaysia, India, Australia, and parts of Pakistan and Sri Lanka, alongside civil-law systems in Japan, Korea, Indonesia, Vietnam, Thailand, and the People’s Republic of China. We have hybrid systems in the Philippines and elsewhere. We have the layered influence of customary, religious, and Sharia-influenced law in various areas. And we have many languages and writing systems. This is the regional inheritance within which any new legal order must be built.

33. Two structural barriers, which have shaped this regional inheritance for decades, are now being dismantled by AI almost simultaneously. The first is the cost of translation. The second is access to comparative law. The ability to engage seriously with the law of a neighbouring jurisdiction, once the preserve of specialists with library and language privileges, is becoming widely available, with reasonable competence, at a keystroke. Both are real and welcome developments.

34. The first thread, then, is liberating. Cross-border practice across our region is becoming easier. Comparative reasoning in our domestic courts is becoming better-informed. Regional dispute resolution is becoming more sophisticated. We have seen, in Singapore, the practical benefits of all this through our work with Indian counterparts, the Sejong dialogue,⁸ recent memoranda of understanding with India and Cambodia,⁹ and the active engagement of the Singapore International Commercial Court. The launch of LawNet 4.0 band the broader work of the Singapore Academy of Law, illustrate one institutional response.¹⁰

35. There is, however, a second thread that runs in tension with the first, and which deserves equal billing. Asian markets remain different from one another, and in some respects they are differing more than they did. To suppose that the dismantling of translation and comparative-law barriers means uncomplicated convergence would be naive.

36. The regulatory landscape is more varied today than it was twelve months ago, and things are moving fast. Vietnam has its ‘Law on Artificial Intelligence’¹¹, South Korea has the ‘Framework Act on Artificial Intelligence’¹² while Japan has the AI Basic Act, the Guidelines for Ensuring Appropriateness published in December last year, and a one-trillion-yen public investment commitment.¹³ Other examples abound, including Singapore’s Model AI Governance Framework.

37. There is divergence of another kind that we should not ignore. AI training data, tooling, and benchmark performance are unevenly distributed across our languages and our legal systems. English-language common-law practice is well served. Civil law in non-English languages is less well served. Minority languages and smaller jurisdictions face the steepest gap. The capability AI offers our region is real. It is not evenly available.

38. There is also a divergence of market scale. The economies that would justify deep investment in legal AI infrastructure are scarce across much of regional practice. Mid-sized firms operating across two or three jurisdictions face a particularly difficult cost calculus. Smaller bars in smaller jurisdictions face a sharper version of the same problem.

39. Without deliberate cooperation, there is a danger that intra-regional capability gaps will widen, and access-to-justice gaps may widen with them. It has been projected that AI could add up to one trillion United States dollars to regional GDP by 2030, but this is accompanied with the warning that the economic and technological divide may deepen.¹⁴

40. The competitive landscape between regional legal-AI hubs is no longer rhetorical. It is becoming visible in concrete launches. The Hong Kong Polytechnic University, in late January this year, launched a cross-border legal AI agent system aiming to provide legal services supporting mainland companies.¹⁵ The Shenzhen Intermediate People’s Court, in June 2024, became the first court systematically to integrate a large language model into judicial reasoning.¹⁶ The South Korean judiciary launched a Judicial Artificial Intelligence Committee in 2025.¹⁷ The ASEAN Guide on AI Governance and Ethics has been promulgated¹⁸. All these initiatives describe an active regional ecosystem, of which Singapore is one node. 

41. That raises the question of reconciling the blurring and the divergence. We must aim to capture the benefits of blurring while addressing the costs of divergence, through deliberate cooperation, judicial, professional, and governmental. Bodies such as ASEAN, the Asian Business Law Institute, APEC, and UNCITRAL matter more because the fragmentation in our part of the world leans us towards divergence rather than convergence. If we can overcome this tendency, Asia has an opportunity here to shape the legal architecture of the AI era. 

42. There is one thing I must emphasise. Convergence is not the same as homogenisation. The legitimacy of our legal systems still rests, ultimately, on local roots, on the history, the languages, the cultures, and the political settlements out of which each of our jurisdictions has emerged. Institutional design remains the deeper task. That task is regional as much as it is national.

VI. Adaptability and Agility

43. What does it mean then for the lawyer or judge in an AI world? Adaptability has to be a habit, a part of daily life. The institutions that fared best in earlier waves of change were those that could undertake a one-off transformation and then settle into a new equilibrium. That model is no longer available to us. The pace of AI development, the speed of regulatory response, and the shifting expectations of those who use legal services together describe an environment in which change is constant.

44. This is not however to recommend merely perpetual motion. What we must do is to respond to the challenges while maintaining our core capabilities and our underlying values, such as empathy, independence, fairness and equity. These are the aspects of legal and judicial work that cannot yet be replaced. With that foundation, there are few measures that institutions and firms can continue to explore to maintain relevance. The first is principled experimentation, that is small bets, fast feedback, and a willingness to be embarrassed by early attempts. The second is discardability, the willingness to abandon sunk expense, outdated processes and obsolete tools when they no longer serve. The third is continued, lifelong, learning and training. 

45. I want to caution, however, against false agility. Movement is not necessarily progress. The history of institutional change in the legal profession is littered with examples of energetic but misdirected reform. 

46. The institutions and organisations that succeed in this environment will shave almost irreconcilable qualities: firstly, humility, that is an acknowledgement that the present generation does not have all the answers; secondly, seriousness, that is a focus on substance; and thirdly confidence, with a belief that even in the age of AI we have a role to play in furthering our institutional values and objectives.

VII. The Road Ahead

47. Two hundred years ago, the founding institutions of Singapore's legal order, and in different ways the founding institutions of legal orders across this region, were established under conditions of profound uncertainty. The choices that were made then, often by people who could not see over the horizon, has had ripples over the two centuries. Thus also, the choices we are making now, about AI, about the structure of the profession, about the work of judging, and about the architecture of regional cooperation, will ripple for many more.

48. What is required of all of us is clarity of sight: clear about what AI can do, candid about what it cannot or should not, and disciplined about the difference.

49. Thank you.