TAJUDIN BIN GULAM RASUL & Anor v SURIAYA BINTE HAJA MOHIDEEN

Case Details

• Citation: [2025] SGHCR 33
• Court: General Division of the High Court (Assistant Registrar)
• Decision Date: 29 September 2025
• Coram: AR Tan Yu Qing
• Case Number: Originating Claim No 125 of 2025; Summons No 1240 of 2025
• Hearing Date(s): 4 June 2025, 22–23 July 2025, 31 July 2025
• Claimants / Plaintiffs: Tajudin bin Gulam Rasul; Mohamed Ghouse s/o Tajudin
• Respondent / Defendant: Suriaya bte Haja Mohideen
• Counsel for Claimants: Lalwani Anil Mangan (DL Law Corporation)
• Counsel for Respondent: Umar Abdullah bin Mazeli (Adel Law LLC)
• Practice Areas: Civil Procedure; Costs; Personal Costs Orders; Professional Conduct; Artificial Intelligence in Legal Practice
• Subject Matter: Citation of a fictitious authority generated by Generative Artificial Intelligence (GenAI) tools and the resulting personal costs consequences for counsel.

Summary

The decision in Tajudin bin Gulam Rasul & Anor v Suriaya binte Haja Mohideen [2025] SGHCR 33 represents a landmark moment in Singapore’s judicial response to the integration of Generative Artificial Intelligence (GenAI) in legal practice. The case primarily concerned an application by the Defendant to set aside a default judgment, but it evolved into a significant inquiry into the professional duties of advocates and solicitors when utilizing AI tools. The central controversy arose when Counsel for the Claimants (CC) cited a fictitious authority in written submissions—a case that did not exist and was later revealed to be a "hallucination" produced by a GenAI tool. This judgment serves as a stern reminder that the duty to the court is non-delegable and that the use of technology does not absolve a practitioner from the fundamental obligation to verify the accuracy and existence of cited authorities.

The court’s analysis focused on the inherent power of the judiciary to supervise the conduct of its officers. While the underlying application to set aside the default judgment was ultimately dismissed on its merits, the court took the proactive step of inviting submissions on whether a personal costs order should be made against CC. This was necessitated by the discovery that CC had relied on a case that was "conspicuously missing" from the Bundle of Authorities and which the Defendant’s counsel (DC) could not locate despite diligent searching. The court emphasized that the citation of a fictitious authority is "wholly impermissible" and strikes at the heart of the administration of justice, which relies on the mutual trust between the Bench and the Bar.

Doctrinally, the court applied the well-established three-stage test for personal costs orders, assessing whether the conduct was improper, unreasonable, or negligent. The court found that CC’s failure to verify the existence of the authority met this threshold. The judgment highlights that even if the legal proposition being advanced is "trite," the act of placing a non-existent authority before the court is a breach of the duty to assist in the administration of justice. The court also considered the broader implications of GenAI, referencing the "Guide on the Use of Generative Artificial Intelligence Tools by Court Users" (the GenAI Guide) and noting that a reasonable advocate does not "blindly rely" on AI tools.

The broader significance of this case lies in its calibration of the "just compensation" stage of the costs analysis. The court held that the gravity of such improper conduct is not merely measured by wasted judicial time or the opposing party's costs, but by the potential "fissure" it creates in the public’s perception of the legal profession. By ordering CC to pay $800 in personal costs to the Defendant, the court signaled that while it would not be purely punitive, it would ensure that the integrity of the justice system is protected and that there is a clear deterrent against the negligent use of AI in court proceedings.

Timeline of Events

1. 8 January 2025: Commencement of the underlying proceedings in Originating Claim No 125 of 2025.
2. 15 April 2024: (Date noted in procedural history regarding prior interactions or filings).
3. 31 May 2025: Preparation of the initial Written Submissions by the Claimants.
4. 1 June 2025: Claimants filed Written Submissions (“CWS”) containing a citation to a fictitious authority to rebut a potential defence under the Moneylenders Act 2008.
5. 2 June 2025: Claimants filed their Bundle of Authorities (“CBOA”), which notably did not include the fictitious authority cited in the CWS.
6. 4 June 2025: Initial hearing date for the substantive application (SUM 1240).
7. 5 June 2025: Court issued directions for parties to ensure all cited authorities were exhibited and to provide pinpoint citations.
8. 18 July 2025: Defendant’s counsel (DC) informed Claimants’ counsel (CC) via email that the fictitious authority cited in the CWS could not be located in any legal database.
9. 21 July 2025: CC filed Amended Written Submissions (“AWS”) and a Supplementary Bundle of Authorities (“CSBOA”) on e-Litigation, replacing the fictitious authority with a legitimate replacement authority.
10. 22 July 2025: Substantive hearing of SUM 1240 commenced. CC admitted that the fictitious authority "did not exist" and was a result of GenAI use.
11. 23 July 2025: Continuation of the hearing on the merits of the setting-aside application.
12. 31 July 2025: The court dismissed the Defendant’s application (SUM 1240) to set aside the default judgment.
13. 6 August 2025: Further proceedings or submissions regarding the issue of personal costs.
14. 29 September 2025: Delivery of the Grounds of Decision regarding the personal costs order against CC.

What Were the Facts of This Case?

The dispute originated from Originating Claim No 125 of 2025 (OC 125), where the Claimants, Tajudin bin Gulam Rasul and Mohamed Ghouse s/o Tajudin, sought legal remedies against the Defendant, Suriaya bte Haja Mohideen. Following the Defendant's failure to file a Notice of Intention to Contest or Not Contest within the prescribed timelines, the Claimants obtained a default judgment. The Defendant subsequently filed Summons No 1240 of 2025 (SUM 1240) seeking to set aside this default judgment, asserting that she had a prima facie defence to the Claimants' claims.

The factual matrix of the costs dispute centered on the Claimants’ Written Submissions (CWS) filed on 1 June 2025. In these submissions, CC anticipated that the Defendant would rely on a defence of illegality under the Moneylenders Act 2008. To counter this, CC argued that the Moneylenders Act was inapplicable to the transaction in question. Specifically, CC averred that the transaction involved isolated or ad hoc loans between acquaintances, which did not constitute "moneylending" unless there was a systematic business of lending for profit. To support this trite legal proposition, CC cited a purported case (the "Fictitious Authority").

The Fictitious Authority was described with a specific case name and citation, appearing as a legitimate precedent. However, when the Claimants filed their Bundle of Authorities (CBOA) on 2 June 2025, the Fictitious Authority was missing. On 5 June 2025, the court issued a general direction to both parties to ensure that all authorities cited in their submissions were included in their respective bundles and that pinpoint citations were provided. This is a standard procedural requirement intended to facilitate the court's review of the law.

The discrepancy was first identified by DC. On 18 July 2025, DC sent an email to CC stating that he had been unable to locate the Fictitious Authority despite searching various legal databases. In response to this notification, CC did not immediately seek leave from the court to amend the submissions. Instead, on the afternoon of 21 July 2025—less than 24 hours before the scheduled hearing—CC filed Amended Written Submissions (AWS) and a Supplementary Bundle of Authorities (CSBOA) on e-Litigation. The AWS replaced the Fictitious Authority with a legitimate case that stood for the same legal proposition.

At the hearing on 22 July 2025, the court queried the late filing of the AWS. CC initially attributed the amendments to "issues with paragraph numbering" and the fact that a case had been "wrongly cited." However, DC clarified that the issue was not a mere typographical error or a wrong citation of an existing case; rather, the case cited in the original CWS simply did not exist. CC eventually conceded that the Fictitious Authority was non-existent. CC explained that the original CWS had been drafted by a junior lawyer in the firm. CC, as the lead counsel, had taken over the matter and only verified the authorities after DC raised the issue. It was subsequently admitted that the Fictitious Authority was a product of a GenAI tool used during the drafting process.

The court proceeded to hear the merits of SUM 1240 and dismissed the Defendant's application on 31 July 2025, finding that no prima facie defence had been established. Despite the dismissal of the application, the court remained concerned about the citation of the Fictitious Authority. The court invited both parties to provide further submissions on whether a personal costs order should be made against CC under the court's inherent jurisdiction and the Legal Profession Act.

What Were the Key Legal Issues?

The primary legal issue before the court was whether a personal costs order, separate from the costs of the substantive application, should be made against CC as a result of the citation of the Fictitious Authority in the CWS. This issue required the court to navigate the intersection of traditional professional duties and the emerging challenges posed by GenAI tools.

The court identified several sub-issues and doctrinal hooks necessary to resolve this matter:

• The Scope of Professional Duty: Whether the duty of an advocate and solicitor to ensure the accuracy of materials placed before the court is non-delegable, particularly when using AI tools or relying on junior staff. This involves an interpretation of the solicitor's role as an "officer of the court" under section 82(1) of the Legal Profession Act 1966.
• The Application of the Ridehalgh Test: Whether CC’s conduct in citing a fictitious authority met the three-stage test for personal costs orders:
  • Stage 1: Was the conduct improper, unreasonable, or negligent?
  • Stage 2: Did the conduct result in unnecessary costs being incurred by the other party?
  • Stage 3: Is it just to order the solicitor to compensate the other party for those costs?
• The Impact of GenAI on Legal Standards: How the "Guide on the Use of Generative Artificial Intelligence Tools by Court Users" (GenAI Guide) informs the standard of care expected of a "reasonable advocate and solicitor."
• The Nature of "Just Compensation": Whether the court, in assessing the quantum of personal costs, should consider broader factors such as the integrity of the justice system and the need for deterrence, rather than merely the mathematical calculation of wasted time.

These issues are critical because they define the boundaries of acceptable practice in a technologically advancing legal landscape. The court had to balance the need to encourage technological adoption with the absolute necessity of maintaining the reliability of legal submissions.

How Did the Court Analyse the Issues?

The court’s analysis began with a fundamental restatement of the role of the advocate. Citing Law Society of Singapore v Udeh Kumar s/o Sethuraju and another matter [2017] 4 SLR 1369, the court emphasized that solicitors are "entrusted with the solemn duty of assisting the court in the administration of justice in Singapore" (at [1]). This duty is not merely aspirational but is a legally binding obligation rooted in the solicitor's status as an officer of the court under section 82(1) of the Legal Profession Act 1966.

The Inherent Power to Impose Personal Costs

The court affirmed its inherent power to impose personal costs on advocates and solicitors, as established in Zhou Tong and others v Public Prosecutor [2010] 4 SLR 534 at [22]. This power is premised on two "practical and ethical considerations" identified in Tan King Hiang v United Engineers (Singapore) Pte Ltd [2005] 3 SLR(R) 529: first, the court's right and duty to supervise the conduct of its officers; and second, the need to penalize conduct that tends to defeat justice. The court noted that a personal costs order is not a sentencing exercise but a principled assessment of the specific conduct in question, citing Syed Suhail bin Syed Zin v Public Prosecutor [2021] 2 SLR 377.

The Three-Stage Ridehalgh Test

The court applied the three-stage test from Ridehalgh v Horsefield [1994] Ch 205, as endorsed by the Singapore Court of Appeal in Tang Liang Hong v Lee Kuan Yew and another and other appeals [1997] 3 SLR(R) 576 at [71].

Stage 1: Improper, Unreasonable, or Negligent Conduct

The court found that CC’s conduct was, at the very least, negligent. The court defined "negligent" in this context as a failure to act with the competence reasonably to be expected of ordinary members of the profession. The court held:

"advocates and solicitors have a professional, non-delegable duty to ensure that all materials placed before the court exist and are accurate." (at [52])

The court rejected any suggestion that the use of GenAI tools or the delegation of drafting to a junior lawyer could mitigate this duty. The court observed that the Fictitious Authority was "conspicuously missing" from the original Bundle of Authorities, which should have served as a "red flag" to CC. The court reasoned that a reasonable solicitor would have verified the existence of the case before signing off on the submissions. The court further noted that the citation of a fictitious authority is "wholly impermissible" because it undermines the adversarial system's reliance on the accuracy of legal citations.

Stage 2: Unnecessary Costs

The court analyzed whether DC had incurred unnecessary costs. DC had spent time searching for the Fictitious Authority across multiple databases and subsequently had to correspond with CC to flag the issue. The court found that these efforts were a direct consequence of CC’s failure to verify the authority. Even though the legal proposition itself was trite, the specific act of citing a non-existent case forced the opposing counsel to expend resources to confirm its non-existence. The court concluded that these costs were "unnecessary" within the meaning of the Ridehalgh test.

Stage 3: Just Compensation

In determining whether it was "just" to make the order, the court looked beyond mere financial loss. The court articulated a broader vision of justice:

"The gravity of improper conduct... is not limited to the wasted judicial time or the unnecessary expenditure by the opposing party... such conduct creates a 'fissure' in the public’s perception of the legal profession and its legitimacy as a custodian of justice." (at [76])

The court emphasized that the integrity of the justice system depends on the "truth-seeking" function of the court, which is compromised when fictitious information is introduced. The court also considered the need for deterrence, particularly given the increasing prevalence of GenAI tools. The court noted that while CC had eventually corrected the error, the correction only occurred after DC had pointed it out. This "reactive" rather than "proactive" correction weighed in favor of making a personal costs order.

The Role of the GenAI Guide

The court integrated the "Guide on the Use of Generative Artificial Intelligence Tools by Court Users" (GenAI Guide) into its analysis. The GenAI Guide stipulates that court users are "personally responsible" for the accuracy of their submissions and must not rely on GenAI to "replace their own independent research." The court held that a reasonable advocate "does not blindly rely on AI [tools] to generate [court documents]" (at [84]). This reinforces the non-delegable nature of the duty of verification.

International Context

The court drew parallels with international developments, noting that the phenomenon of "hallucinations" in AI-generated legal work has been observed in the UK, US, Canada, and Australia. The court cited Ayinde v London Borough of Haringey [2025] EWHC 1383 (Admin) and Luck v Secretary, Services Australia [2025] FCAFC 26 as examples where courts have had to address the citation of fictitious authorities. By aligning with these jurisdictions, the court signaled that Singapore’s approach is consistent with global standards for professional conduct in the digital age.

What Was the Outcome?

The court’s decision resulted in two distinct outcomes: one regarding the substantive application and one regarding the personal conduct of counsel. On the merits of the application (SUM 1240), the court dismissed the Defendant’s request to set aside the default judgment. The court found that the Defendant had failed to demonstrate a prima facie defence that would warrant reopening the case. Consequently, the default judgment remained in force, and the Defendant was ordered to pay the costs of the application to the Claimants.

However, with respect to the professional conduct issue, the court exercised its inherent jurisdiction to impose a personal costs order against CC. The court’s final order was as follows:

"Accordingly, I ordered CC to personally pay the Defendant costs in the sum of $800, being the costs of and incidental to his citation of the Fictitious Authority in the CWS, within 14 days of my decision." (at [99])

The court calibrated the quantum of $800 by considering the time DC spent searching for the case and the subsequent correspondence. The court noted that while the amount was relatively modest, it was intended to be a "principled assessment" that balanced the need for compensation with the fact that the error was eventually corrected. The court also ordered that CC could not recover this sum from the Claimants, ensuring the burden fell solely on the solicitor responsible for the oversight.

Furthermore, the court took the unusual step of omitting the name and case number of the Fictitious Authority from the Grounds of Decision. This was done to prevent the further propagation of false information, following the approach taken by the Federal Court of Australia in Luck v Secretary, Services Australia [2025] FCAFC 26 at [14]. The court also directed that both counsel provide a copy of the court’s directions to their respective clients, ensuring transparency regarding the conduct that led to the personal costs order.

The court concluded by reiterating that the use of GenAI is not prohibited but must be accompanied by rigorous human oversight. The outcome serves as a clear precedent: the court will not tolerate the introduction of fictitious materials, and the costs of identifying and correcting such errors will be borne personally by the negligent practitioner.

Why Does This Case Matter?

This case is of paramount importance to the Singapore legal landscape for several reasons. First, it provides the first detailed judicial analysis of the professional consequences of "AI hallucinations" in court submissions. While the legal profession has been eager to adopt GenAI for efficiency, this judgment sets a clear boundary: efficiency cannot come at the expense of accuracy. The court’s insistence that the duty of verification is "non-delegable" means that senior practitioners cannot shield themselves from liability by blaming junior staff or the software they use.

Second, the decision reinforces the "officer of the court" doctrine in a modern context. It clarifies that the relationship between the Bar and the Bench is built on a presumption of honesty and diligence. When a solicitor cites a case, the court should be able to assume that the case exists and has been read. The citation of a fictitious authority breaks this "social contract" of the courtroom. The court’s discussion of the "fissure" in public perception (at [76]) suggests that the judiciary views such conduct as a threat to the very legitimacy of the legal system.

Third, the judgment provides a practical application of the Ridehalgh test to technological negligence. It demonstrates that "negligence" in the 21st century includes the failure to verify AI-generated output. This expands the traditional understanding of professional negligence to include "technological incompetence." Practitioners must now recognize that the "reasonable solicitor" standard includes a baseline level of AI literacy and a skeptical approach to AI-generated citations.

Fourth, the case highlights the importance of the "GenAI Guide" issued by the Singapore courts. By citing the Guide, the court has elevated it from a set of recommendations to a standard of conduct that will be used to assess negligence in personal costs applications. This gives the Guide significant "teeth" and makes it essential reading for every practitioner in Singapore.

Fifth, the decision aligns Singapore with other leading common law jurisdictions. By referencing cases from the UK and Australia, the court has ensured that Singapore remains at the forefront of global legal thought regarding AI regulation. This consistency is vital for Singapore’s reputation as a leading international legal hub.

Finally, the modest quantum of the costs order ($800) suggests that the court’s primary goal is not to be draconian but to be corrective. The judgment is a "shot across the bows" for the profession. It signals that while the court understands that mistakes happen, it will not allow the costs of those mistakes to be shifted to the opposing party or ignored. It encourages a culture of "proactive correction" and rigorous internal quality control.

Practice Pointers

• Non-Delegable Duty of Verification: Advocates and solicitors must personally verify the existence, accuracy, and relevance of every authority cited in their submissions. This duty cannot be delegated to junior lawyers, paralegals, or AI tools.
• The "Bundle of Authorities" Check: The process of compiling a Bundle of Authorities (BOA) should serve as a final audit. If a case cited in the submissions cannot be found for the BOA, it is a critical "red flag" that must be investigated immediately before filing.
• Proactive Disclosure: If a mistake is discovered, counsel should proactively inform the court and the opposing party at the earliest opportunity. Reactive corrections—those made only after the other side points out the error—are viewed less favorably by the court.
• Supervision of Junior Staff: Senior practitioners must implement robust systems for supervising junior staff who use GenAI tools. "Signing off" on a document drafted by a junior involves a personal representation to the court that the contents are accurate.
• Adherence to the GenAI Guide: All practitioners must familiarize themselves with the "Guide on the Use of Generative Artificial Intelligence Tools by Court Users." Compliance with the Guide is a key factor in determining whether a solicitor has acted "reasonably."
• Independent Research: GenAI should be used as a starting point or a brainstorming tool, not a replacement for traditional legal research databases (like LawNet or Westlaw). Every AI-suggested citation must be "cross-checked" against an official reporter.
• Transparency with Clients: When a personal costs order is made, counsel may be required to provide a copy of the court's directions to their client. This can have significant reputational consequences and impact the solicitor-client relationship.
• Pinpoint Citations: Always provide pinpoint citations. The inability to find a specific paragraph or page number in a cited case is often the first sign that the authority may be fictitious or "hallucinated."

Subsequent Treatment

As of the date of this decision, [2025] SGHCR 33 stands as a foundational authority in Singapore regarding the intersection of GenAI and professional conduct. It applies the Ridehalgh three-stage test to the specific context of AI-generated fictitious authorities. The ratio of the case establishes that the court has the inherent power to impose personal costs on solicitors who fail in their non-delegable duty to verify materials placed before the court. While it is a decision of the Assistant Registrar, its detailed reasoning and reliance on Court of Appeal precedents (such as Tang Liang Hong and Zhou Tong) suggest it will be highly persuasive in future disputes involving technological negligence. The case has already been noted for its alignment with international trends in the UK and Australia, reinforcing a global judicial consensus on the need for human oversight in AI-assisted legal practice.

Legislation Referenced

• Legal Profession Act 1966 (2020 Rev Ed): Section 82(1) (Solicitors as officers of the court).
• Moneylenders Act 2008 (2020 Rev Ed): Interpreted in the context of the underlying dispute regarding the definition of "moneylending."
• Rules of Court 2021: Order 21 Rule 6 (referenced in the context of costs and the court's general powers).

Cases Cited

• Applied:
  • Tang Liang Hong v Lee Kuan Yew and another and other appeals [1997] 3 SLR(R) 576 (Endorsing the Ridehalgh test).
  • Ridehalgh v Horsefield [1994] Ch 205 (The three-stage test for personal costs).
  • Zhou Tong and others v Public Prosecutor [2010] 4 SLR 534 (Inherent power to impose personal costs).
  • Tan King Hiang v United Engineers (Singapore) Pte Ltd [2005] 3 SLR(R) 529 (Practical and ethical considerations for personal costs).
  • Luck v Secretary, Services Australia [2025] FCAFC 26 (Approach to fictitious AI authorities).
• Referred to:
  • Law Society of Singapore v Udeh Kumar s/o Sethuraju and another matter [2017] 4 SLR 1369 (Duty to assist in the administration of justice).
  • Ayinde v London Borough of Haringey [2025] EWHC 1383 (Admin) (International trends in AI hallucinations).
  • Syed Suhail bin Syed Zin v Public Prosecutor [2021] 2 SLR 377 (Nature of personal costs inquiry).
  • Lee Tat Development Pte Ltd v Management Corporation Strata Title Plan No 301 [2018] 2 SLR 866 (Principle of stare decisis).
  • Toh Siew Kee v Ho Ah Lam Ferrocement (Pte) Ltd and others [2013] 3 SLR 284 (Role of decided cases).
  • Ayinde v London Borough of Haringey [2025] EWHC 1040 (Admin) (Specific findings on AI negligence).

Source Documents

• Original Judgment at eLitigation
• Judgment PDF (CDN)