Tan Hai Peng Micheal and another (as the executors of the e state of Tan Thuan Teck, deceased) v Tan Cheong Joo and another and other matters [2026] SGHC 49

Generative artificial intelligence (“GenAI”) is reshaping the legal profession in a manner that can fairly be described as nothing less than a sea change. While the technological advances that GenAI has engendered is not in itself unwelcome, its integration into legal practice demands a commensurate evolution in professional responsibility. The very standards and ethical foundations of the profession depend on it. The responsibility that lies on the shoulders of every advocate and solicitor utilising GenAI has been repeatedly underscored in both the professional guidelines governing legal practice, and in judicial authorities. It is again re-emphasised in this judgment.

This case brought into sharp focus the timely issue of the use (or rather, misuse) of artificial intelligence (“AI”) tools in the conduct of court proceedings, and this judgment is the sequel to my earlier decision in Tan Hai Peng Micheal v Tan Cheong Joo [2025] SGHC 217 (the “First Judgment”). In this judgment, I will continue to adopt the same abbreviations I used in the First Judgment, particularly those in the section where I had discussed the issue concerning the citation of fictitious authorities.

Background Facts

Facts in the First Judgment

The background facts to this case are fully set out in the First Judgment. I do not propose to rehearse them here, but in brief, the main facts are as follows. The claim arose from several loans made by the late Tan Thuan Teck (“TTT”) to the defendants. The defendants are (a) four brothers: Tan Cheong Joo, Tan Seong Kok, Tan Siong Tiew, and Tan Siong Lim (collectively, the “Defendants”); and (b) Fong Tat Holding Co Pte Ltd, a company in which the brothers are directors and/or shareholders.

TTT’s sons, who are also the co-executors of his estate (the “Claimants”), commenced actions to recover sums they allege remain outstanding under the loans. There were three actions brought, namely HC/OC 381/2023, HC/OC 382/2023, and HC/OC 201/2024 (the “Suits”). The Suits were informally consolidated and I heard the trial last year.

In the First Judgment, I found that the Defendants failed to establish any of their defences and that the claim was made out. Accordingly, I granted the Claimants judgment in the Suits.

Facts material to the present judgment

Of central relevance to the present judgment is the fact that, in the Defendants’ closing submissions, Mr Goh Peck San (“Mr Goh”), counsel for the Defendants, cited two fictitious authorities at paragraphs 67 and 68, abbreviated in the First Judgment as the “Relevant Paragraphs” (the “Fictitious Authorities Issue”) – see the First Judgment at [95]. The Relevant Paragraphs cited two “cases”, which I shall refer to as “Case A” and “Case B” respectively (collectively, the “Fictitious Authorities”). The fact of the Fictitious Authorities being cited was discovered by the Claimants’ counsel, and highlighted by them in the Claimants’ reply submissions. As I noted in the First Judgment at [100], there were reasonable grounds to suspect that the citations were likely to have been generated by an AI tool.

Chronology of Events

I first set out below the relevant background facts on the Fictitious Authorities Issue, following the court’s investigations and compilation of the relevant information.

Counsel on record for the Defendants was Mr Goh. Mr Goh had at some point engaged another solicitor, Mr Sidhu, to assist him in carrying out research for and in the drafting of the Defendants’ closing submissions.

On 4 August 2025, the Claimants brought the Fictitious Authorities Issue to the court’s attention in their reply submissions. The Claimants highlighted that although the neutral citation and the name ascribed to Case A, individually, corresponded to real cases, the neutral citation belonged to an entirely different case – thus, when put together, Case A with that name and that neutral citation did not exist. As regards Case B, both the name of the case and the neutral citation were entirely fictitious. In addition, neither case (whether by neutral citation or by case name) supported the legal propositions that the Defendants claimed they established.

Following these allegations, the court registry directed Mr Goh to respond and confirm if he agreed with the allegations raised by the Claimant. After a series of communications with the court, on 10 September 2025, Mr Goh eventually acknowledged and agreed with the Claimants’ allegations, and explained that the Fictitious Authorities had been provided by a fellow solicitor whom he had engaged to assist with research – the fellow solicitor was not identified in Mr Goh’s response. He further stated that he was unaware of which AI tool had been used, expressed his regret, and sought the court's indulgence.

As Mr Goh’s response only raised more questions, the court deemed it necessary to seek further clarification. On 11 September 2025, the Defendants’ counsel was directed (a) to identify the solicitor engaged to assist with the research for and in the drafting of the Defendants’ closing submissions, including the law practice(s) in which the said solicitor practised; and (b) to confirm whether the research underpinning the Relevant Paragraphs was generated using an AI tool and, if so, which tool, and what steps, if any, were taken to verify the accuracy of the output. Both Mr Goh and the solicitor he engaged were directed to provide their responses to the court and were reminded to provide full and forthright responses.

On 18 September 2025, Mr Goh informed the court that Mr Sidhu had been instructed by Mr Goh to assist in research and in drafting the submissions. Mr Goh’s response also enclosed a letter from Mr Sidhu dated 17 September 2025. In that letter, Mr Sidhu stated that the initial research and draft of the closing submissions were prepared by a paralegal (unnamed) who had since left the firm, that Mr Sidhu was unaware whether any AI tools had been used by the paralegal, and that the draft submissions and cases were reviewed by Mr Sidhu and transmitted to Mr Goh. Mr Sidhu conceded that he had overlooked the Fictitious Authorities. Although he later realised that the two cases he had sent to Mr Goh were not on point for the submissions made in the Relevant Paragraphs and informed Mr Goh that another case could not be located on LawNet and should be removed, he forgot to inform Mr Goh that the two cases he had sent were not on point. As far as the issues concerning Case A and Case B were concerned, he overlooked them. Mr Sidhu further stated that despite Case A and Case B being fictitious, his firm maintains an internal database containing authorities that do support the legal principles set out at the Relevant Paragraphs. Mr Sidhu also apologised to the court and to Mr Goh.

Parties’ arguments

Issues

Accordingly, in this judgment, I determine these questions in the following order:

(a) on the Fictitious Authorities Issue, whether it warrants a personal costs order against Mr Goh and/or Mr Sidhu;

(b) the quantum of any personal costs order(s) made against Mr Goh and/or Mr Sidhu; and

(c) the quantum of costs for the Suits that the Claimants are entitled to recover.

As I stated in the First Judgment at [95], to prevent further dissemination of false information, I refrain from repeating the actual citations of the Fictitious Authorities in this judgment. As abbreviated above at [6], they will simply be referred to as “Case A” and “Case B”.

The Fictitious Authorities Issue

To begin, I observe that the troubling issue of counsel relying on AI-generated case authorities, without verification of their existence or accuracy, is attracting mounting concern in a number of jurisdictions. In less than a year after Tajudin was decided, more cases have emerged beyond those mentioned in Tajudin. For example, both Tajudin and the First Judgment cited the English decision in Ayinde v London Borough of Haringey [2025] 1 WLR 5147 (“Ayinde”) – the judgment in Ayinde was published on 6 June 2025. On 2 July 2025, less than a month after the Ayinde judgment, the Federal Court of Australia handed down its judgment in Murray (on behalf of the Wamba Wemba Native Title Claim Group) v Victoria [2025] FCA 731 (“Wamba Wemba”), where the court observed as follows at [11]–[12]:

Shortly thereafter, on 15 August 2025, the Federal Circuit and Family Court of Australia handed down judgment in JNE24 v Minister for Immigration & Citizenship [2025] FedCFamC2G 1314 (“JNE24”). JNE24 involved a case where the applicant’s counsel similarly relied on fictitious cases in his submissions to the court as a result of AI-generated research. At [24]–[25] of JNE24, Judge Gerrard made the following pointed remarks on the dangers of AI-generated legal research:

And shortly thereafter, on 12 September 2025, the Federal Circuit and Family Court of Australia handed down judgment in a third case from Australia – Helmold v Mariya (No 2) [2025] FedCFamC1A 163 (“Helmold”). In Helmold, the appellant’s counsel admitted to using AI to assist in the preparation of the appellant’s Notice of Appeal and Summary of Argument. At [7], the court reproduced the observations of Dame Victoria Sharp in Ayinde regarding the:

In Singapore, likewise, Forbes Monaco APAC v Kawajiri Seiji [2026] SGHC 16 (“Forbes”), decided in January this year, represents yet another matter brought before the Singapore Courts involving the citation of fictitious authorities. The decision in Forbes follows in the wake of existing cases in Singapore such as Tajudin and Landscape Engineering Pte Ltd v Dot Safety Solutions Pte Ltd [2025] SGHC 214 (“Landscape Engineering”). While Forbes and Landscape Engineering concerned self-represented litigants rather than parties represented by counsel, all of the cases referred to above nonetheless demonstrate the prevalence of the problem.

I have referred to and reproduced excerpts from the Australian cases referred to above (at [26]–[28]) to underscore the seriousness of the matter – it strikes at the very heart of the proper administration of justice, and has the potential to severely undermine it. It can be seen from the cases surveyed in Tajudin and those I have cited above that across jurisdictions, courts are increasingly encountering cases in which either legal submissions have been prepared with the use of AI tools or such tools have been relied on for legal research and which have “hallucinated” plausible sounding but entirely fabricated legal authorities.

In my view, this phenomenon, if left unchecked, poses a serious threat to the administration of justice in Singapore, which relies heavily on lawyers to “ensure that the evidence and arguments adduced before the court are skilfully, accurately and fairly presented” [emphasis added] (Attorney-General v Shahira Banu d/o Khaja Moinudeen [2024] 4 SLR 1324 at [60]). And as also noted in Ayinde (at [5]), “the administration of justice depends upon the court being able to rely without question on the integrity of those who appear before it and on their professionalism in only making submissions which can properly be supported” [emphasis added]. The provision of fictitious authorities to the court wastes precious judicial time and resources at best, if the error is discovered in time. At worst, it may lead to an erroneous decision made in mistaken reliance on such fictitious authorities, resulting not only in potential embarrassment but more importantly, a potential miscarriage of justice.

To be clear, there is no inherent objection in using AI tools per se – see Supreme Court Registrar’s Circular No 1 of 2024, Guide on the Use of Generative Artificial Intelligence Tools by Court Users (“AI Circular”), at para 3(1); Tajudin at [40(a)]). In fact, it is beyond argument that AI tools are here to stay and will become an increasingly prevalent feature of our legal ecosystem. As observed by Judge Gerrard in JNE24 at [25] (see [27] above), “it is not the initial reliance on AI that constitutes the vice in such matters”. However, AI tools can only ever play the part of the handmaiden. The fundamental point remains that lawyers bear ultimate responsibility for ensuring that all materials placed before the court – including case authorities – are, without exception, true and accurate, and crucially, verified by them to be so. In view of this inescapable duty as officers of the court, it is immaterial whether the fictitious authorities were generated by an AI tool or by any other means. It goes without saying that a lawyer practising at the Singapore Bar who permits fictitious authorities to be submitted to the court risks contravening, among others, the AI Circular (in a case where AI is used), the Supreme Court Practice Directions, and the Legal Profession (Professional Conduct) Rules 2015 (see, generally, Tajudin at [51]–[59]).

Whether Mr Sidhu can also be held responsible by way of a personal costs order

Applicable framework

(a) first, the court determines whether the advocate and solicitor has acted improperly, unreasonably, or negligently (“Stage 1”);

(b) second, if so, whether such conduct has caused the other party to incur unnecessary costs (“Stage 2”); and

(c) third, if again answered affirmatively, whether it is in all the circumstances just to order the advocate and solicitor to compensate the other party in whole or in part for the costs incurred (“Stage 3”).

Stage 1: Whether Mr Goh and/or Mr Sidhu acted improperly, unreasonably, or negligently

As noted in Tajudin (at [84]–[85]), not only is it unreasonable for a solicitor to be blindly reliant on AI-generated authorities, if the solicitor fails to verify the accuracy of all cases put before the court, such failure would also amount to negligent conduct in failing to act with the competence reasonably expected of solicitors. But it is not just the “unreasonable” or “negligent” limbs in Ridehalgh that I am concerned with and wish to address. Rather, I turn to the fact that Mr Goh’s conduct was, in my view, clearly improper.

Ridehalgh defines improper conduct as follows (at 232–233, and cited in Tajudin at [49]):

Several facts, when considered together, are particularly striking in this case. Most notably, Mr Goh failed to exercise due care and diligence in verifying the existence of the Fictitious Authorities. Notwithstanding the explanations provided, it is still not apparent who took on the ultimate responsibility for finalising the Defendants’ closing submissions prior to filing them. In the absence of any contrary indication, it must be assumed that Mr Goh, as the counsel on record for the Defendants, undertook that responsibility. In doing so, he had a non-delegable duty to ensure that the submissions were complete, accurate, and properly verified. By appending his signature to the final version of the submissions that were filed and served, he thereby accepted personal responsibility for all material submitted to the court.

Further, it would have been manifestly obvious, or ought reasonably to have been so, to any competent solicitor that the bundle of authorities accompanying the submissions was incomplete. It should, and must, have been plainly evident to Mr Goh that Case A and Case B were not included in the bundle of authorities that was to accompany the closing submissions – a reasonably diligent solicitor would have picked up on this as a potential red flag. That Mr Goh took no steps to address this glaring omission is a clear indication that his conduct fell short of that expected of an officer of the court. The proper and expected course of action would have been for Mr Goh to spend the time to track down and verify the existence of the missing authorities. If the missing authorities could not be found, Mr Goh should have removed all references to them from the submissions altogether and to replace those references with actual cases on the point (verified to exist), and for those cases to be included in the bundle of authorities – these are not extraordinary steps that the court expects from counsel, but basic ones.

But leaving aside Mr Goh as counsel on record, Mr Sidhu’s role and conduct in the Fictitious Authorities Issue was also improper. Mr Sidhu acknowledged in correspondence that he “overlooked” the Fictitious Authorities, although he realised that the two cases he did in fact send to Mr Goh were not on point. Even if the matter had not arisen to his attention, he ought to have recognised that a paralegal is not qualified to prepare written submissions independently, and that proper supervision and the verification of all authorities was his ultimate responsibility. To cross-check that the cases cited in the Relevant Paragraphs actually existed was not an onerous task. Yet, despite having read the draft submissions “about twice”, the Fictitious Authorities referred to in the Relevant Paragraphs were not picked up by Mr Sidhu.

Mr Sidhu further asserted that his firm maintains an internal database containing authorities that support the legal principles set out in the Relevant Paragraphs – implying that his error was citing incorrect authorities in support of otherwise correct propositions. This argument is, in my view, without merit and misses the point. Mr Sidhu’s duty as a solicitor and officer of the court was to ensure that every authority referred to and relied upon in the closing submissions that he was tasked to prepare was verified to actually exist and confirmed to be accurate. Of particular gravity is that Case A could, in fact, be mistaken to be a real case, in the sense that the neutral citation and the case name, if each were viewed in isolation, actually existed.

Finally, I found it extraordinary that between Mr Goh and Mr Sidhu, the issue fell between two stools and neither solicitor took the time or trouble of actually verifying the existence and veracity of the Fictitious Authorities. As I indicated above, for Case A, the neutral citation and the case name were individually correct, but when paired together formed an entirely non-existent case. Case B was arguably more egregious – it was entirely fictitious. Neither the neutral citation nor the case name, even if individually searched in a search bar of an internet browser, or a research tool such as LawNet, would have turned up an existing case.

That said, I accept that unlike counsel in Tajudin, Mr Goh and Mr Sidhu did not seek to evade responsibility. Their acceptance of the Claimants’ allegations and their apologies are duly noted, and I accept that their conduct was not intentional. However, their remorse does not detract from the inherent seriousness of their conduct.

What has been a disconcerting feature of this saga is that despite multiple letters from the court, Mr Goh and Mr Sidhu remain unable to provide confirmation on whether the Fictitious Cases had been included in the closing submissions as a result of the use of an AI tool to prepare a draft of the submissions. Even after all their explanations, no such confirmation was forthcoming from either Mr Goh or Mr Sidhu. As I observed in the First Judgment, it was plainly evident that such a tool was most likely employed, and hence the hallucination in terms of the Fictitious Authorities. The fact that neither Mr Goh nor Mr Sidhu could shed any light on this question demonstrated to me the sheer lack of diligence displayed. All that was before me was Mr Sidhu’s assertion that the paralegal who prepared the draft submissions was no longer with the firm and uncontactable, and that he had previously instructed his staff not to utilise such tools. But the fact that the Fictitious Authorities were included in the draft submissions without being picked up by Mr Sidhu (and, it would appear to me, with no inkling whether any AI tools had in fact been utilised by the paralegal) indicates a lax system of control and supervision.

In my view, and bearing in mind the points raised above, this is a clear case that warrants the solicitors responsible being sanctioned by the court.

Stage 2: Whether Mr Goh’s and/or Mr Sidhu’s conduct caused the Claimants to incur unnecessary costs

Stage 3: Whether it is in all the circumstances just to order Mr Goh and/or Mr Sidhu to compensate the Claimants for the whole or any part of the relevant costs

On the issue of just compensation, the Claimants submit that the Fictitious Authorities were cited in support of a central plank of the Defendants’ case – that the Claimants (and not the Defendants) bore the burden of proving that TTT was an excluded moneylender under the MLA. Given that the court had expressly directed the parties to address the issue of burden of proof in their closing submissions, the Fictitious Authorities had the potential to materially affect the court’s reasoning and, consequently, the outcome of the case.

The Claimants also submit that, notwithstanding the need for immediate, full and truthful explanations (Tajudin at [72(c)]; Ayinde at [24]), the responses provided by Mr Goh and Mr Sidhu have been unclear and gaps still remain.

Mr Goh and Mr Sidhu did not seriously dispute these contentions.

In Tajudin at [65], three principles were identified in considering just compensation: (a) fairness and proportionality; (b) the preservation of the integrity of the justice system; and (c) deterrence. In respect of (a), relevant considerations include the nature and impact of the solicitors’ conduct, the prejudice suffered by the opposing party, and the egregiousness of the conduct (at [68]).

Whether Mr Sidhu should be permitted to bear Mr Goh’s share of personal costs

Quantum of costs

Personal costs orders

(a) First, and as a general point, when imposing personal costs orders, the court’s powers are not fettered or guided by any “tariff” – each case depends on its particular facts. As I have explained above, the citation of fictitious authorities here carried a significantly greater potential to affect the outcome of the case as they related to a central issue in the case. In Tajudin, it concerned a peripheral issue.

(b) Second, Tajudin concerned an application to set aside a default judgment, whereas the present matter involved a trial of claims across three separate suits. This is not to suggest that misconduct in smaller matters is in any way excusable or trivial. Rather, the point is that I am not constrained to regard $800 as representing any sort of benchmark.

Finally, in Tajudin at [100], having regard to the seriousness of the matter, AR Tan directed that counsel for both parties furnish their respective clients with copies of her directions, to underscore the seriousness with which the court views such conduct by solicitors.

Costs of the Suits

(a) efforts made by the parties at amicable resolution;

(b) the complexity of the case and the difficulty or novelty of the questions involved;

(c) the skill, specialised knowledge and responsibility required of, and the time and labour expended by, the solicitor;

(d) the urgency and importance of the action to the parties;

(e) the number of solicitors involved in the case for each party;

(f) the conduct of the parties;

(g) the principle of proportionality; and

(h) the stage at which the proceedings were concluded.

(a) Pre-trial costs: The Claimants submit that a sum of $50,000, $60,000 and $50,000 for HC/OC 201/2024, HC/OC 381/2023, and HC/OC 382/2023 respectively would be appropriate before applying a 20% discount to account for overlap – the net figures claimed for pre-trial costs are $40,000, $48,000 and $40,000 respectively. As justification for the figures claimed, the Claimants refer to the range in Appendix G of $25,000-$70,000 per suit for pre-trial costs, the Claimants’ three-tiered legal strategy, the heavily factual nature of the dispute requiring extensive evidential work, and the need for six tailored affidavits of evidence-in-chief due to each suit’s unique circumstances.

(b) Trial costs: The Claimants claim $108,000 for six days of trial across the Suits, pegged at $6,000 per day of trial for each of the Suits. The Claimants referred to Appendix G which provided a range of $6,000-$16,000 for each day of trial.

(c) Post-trial costs: The Claimants claim a base sum of $15,000 for each of the Suits, taking reference from Appendix G which indicates a sum of “up to $30,000” for post-trial work. Accounting for overlap, the Claimants apply a 20% discount and seek a net figure of $36,000 in total for post-trial work (or $12,000 per suit).

In my view, the sums advanced by the Claimants exceed what are reasonable and proportionate.

Disbursements

With respect to disbursements, the Claimants provided a breakdown of the amounts claimed but without any supporting documents. In any event, both parties agree that I should order that disbursements be agreed in the first instance, and failing agreement for the court to fix the disbursements.

Accordingly, I direct that the parties are to endeavour to agree the disbursements for the Suits within three weeks from the date of this judgment. Should the parties be unable to reach an agreement, the Claimants’ solicitors are to write to the court to seek directions for the court to fix the disbursements.

Conclusion

For the foregoing reasons, these are my orders:

(a) costs of the Suits are fixed at the global sum of $180,000 (excluding disbursements), payable to the Claimants;

(b) the global sum of $180,000 is split equally across each suit. Thus, the costs of each suit are fixed at $60,000, payable to the Claimants by the Defendants in each suit jointly and severally;

(c) a sum of $5,000 is ordered as costs against Mr Goh personally in respect of the Fictitious Authorities Issue;

(d) a sum of $5,000 is ordered as costs against Mr Sidhu personally in respect of the Fictitious Authorities Issue;

(e) the global sum of $180,000 ordered at [82(a)] above includes the personal costs ordered above at [82(c)] and [82(d)];

(f) accordingly, the Defendants are, collectively, to pay the Claimants the sum of $170,000 as costs of the Suits while Mr Goh and Mr Sidhu are to each personally pay the Claimants $5,000;

(g) interest on the costs ordered is payable at the rate of 5.33% per annum from the date of this judgment to the date of payment; and

(h) disbursements for the Suits are to be agreed between the parties within three weeks from the date of this judgment, failing which the Claimants are to write to the court for directions for the disbursements to be fixed by the court.

In closing, I offer a word of advice, or perhaps more importantly, caution to the Bar. I hope this judgment leaves no member of the Singapore Bar in any doubt over the critical importance of (a) verifying the authenticity and accuracy of all materials submitted to the court in any proceedings before it, and (b) familiarising themselves with the rules and guidance on the proper use of GenAI in the course of their work (see [32] above). The consequences of submitting fictitious authorities to the court can be severe, not just for the adjudication of the case in question but the administration of justice as a whole and the public image of the legal profession. Adopting the sagacious comments of Dame Victoria Sharp in Ayinde at [9], it is imperative, perhaps now more than ever, that every advocate and solicitor practicing at the Singapore Bar “(whenever and wherever they were qualified to do so) understands and complies with their professional and ethical obligations and their duties to the court if using artificial intelligence”.