ADGM Court Sanctions Law Firm for AI-Generated Hallucinations in Court Filings, Emphasizing Verification Duties

Introduction

In a landmark ruling that underscores the perils of artificial intelligence in legal practice, the Abu Dhabi Global Market (ADGM) Court of First Instance has ordered MIO Legal Consultants LLP to pay AED 282,508 (approximately ₹63.3 lakh) in wasted costs to Arabyads Holding Limited. Delivered by Justice Paul Heath KC on December 18, 2025, the judgment ([2025] ADGMCFI 0032) criticizes the firm's use of unverified AI tools in preparing a defense, resulting in fictitious and mis-cited legal authorities. This decision, arising from a commercial dispute over an employee's alleged breach of a share option plan valued at over USD 11.2 million, serves as a cautionary tale for lawyers worldwide. While acknowledging AI's potential benefits, the court stressed that technological reliance does not absolve professionals of their core duties to verify information and avoid misleading the court. The case highlights growing concerns about AI "hallucinations"—fabricated outputs—in legal research, as noted in the court's analysis of empirical studies and international precedents.

The bench, comprising Justice Paul Heath KC, addressed multiple costs applications stemming from procedural missteps in the underlying claim by Arabyads against former employee Gulrez Alam Marghoob Alam. MIO, which represented Alam initially, faced scrutiny for filing a prolix 233-page defense riddled with errors indicative of inadequate AI oversight. This ruling not only compensates the claimant but also punishes professional lapses, potentially influencing AI adoption policies in law firms globally.

Case Background

The dispute originated from a commercial claim filed by Arabyads Holding Limited, an ADGM-incorporated private company, against its former employee, Gulrez Alam Marghoob Alam, on April 18, 2025. Arabyads alleged that Alam breached an Employee Share Option Plan (ESOP Plan) and a related Grant Agreement, designed to incentivize employees through proprietary interests in the company's growth. Specifically, Arabyads sought damages, declarations that Alam was a "Bad Leaver," expiration of vested restricted shares, and specific performance to transfer shares under a call option. The stakes were high, with the total equity value exceeding USD 11.2 million, including additional vested tranches by April 2024.

Alam, employed by Arabyads' subsidiary Arabyads FZ LLC, instructed MIO Legal Consultants LLP to represent him shortly after service of the claim on April 29, 2025. MIO filed a Transfer Application on May 21, 2025, seeking to move the case to the ADGM Employment Division; a lengthy Defense on June 3, 2025; and a Joinder Application on June 4, 2025, to add Arabyads FZ as a party for a potential counterclaim.

The Defense, spanning 47 pages of narrative (327 paragraphs) plus exhibits, drew immediate criticism from Arabyads for its prolixity and inclusion of non-existent or misapplied authorities. Arabyads responded with a June 17, 2025, application for an "unless" order requiring repleading. Hearings on these applications occurred on July 3, 2025, but Alam abruptly changed representation the day prior, withdrawing the Transfer and Joinder Applications and agreeing to amend the Defense.

An Amended Defense was filed by July 28, 2025, but a proposed Counterclaim was not due to unpaid filing fees and was later abandoned. Post-hearing, Arabyads pursued wasted costs against MIO, alleging AI misuse, while Alam sought similar relief against his former firm. The court directed MIO to explain its research processes, revealing reliance on AI amid resource constraints. Mediation in October 2025 failed, leading to the December judgment amid trial preparations set for June 2026.

The core legal questions revolved around: (1) whether MIO's conduct warranted a wasted costs order under ADGM regulations; (2) the threshold for lawyer liability in AI-assisted research; and (3) assessment of costs on indemnity versus standard basis across applications.

Arguments Presented

Arabyads, represented by counsel Sajid Suleman of Matouk Bassiouny, argued that MIO breached professional duties by filing a Defense with "hallmarks of hallucinatory results" from unverified AI tools, contrary to ADGM Court Rules of Conduct 2016. They highlighted over 20 problematic citations: fictitious cases like "Allied Dunbar [1985]" and "Johnston v Moreton [2014] EWHC 1219 (Ch)"; mis-citations such as "Halifax Life Ltd v DLA Piper LLP [2008]" (actual 2009 Scottish case); and misapplications, e.g., Caparo Industries Plc v Dickman [1990] 2 AC 605 for causation in a context unrelated to its tort duty of care focus. Arabyads claimed this forced extensive review, incurring AED 785,581 in costs, and sought indemnity costs against MIO and Alam jointly for the Defense Application, plus individual claims for Transfer (AED 102,680), Joinder (AED 232,526.50), and Counterclaim (AED 167,866.50) applications.

Arabyads invoked ADGM's wasted costs jurisdiction under section 49 of the Courts Regulations 2015 and Rule 203 of the Court Procedure Rules 2016, asserting MIO's conduct was "unreasonable or improper," triggering mandatory indemnity costs. They dismissed MIO's excuses—time pressures, non-payment by Alam, and lack of UK counsel—as irrelevant to duties owed to the court, emphasizing AI's inherent risks as per studies like Magesh & Ors (2025) on LLM hallucinations.

MIO, through witness Silsy Samuel and Karim Yassine, denied intentional misconduct, claiming AI was used only for "formatting, paragraph structuring, and editing," with research via BAILII, Supreme Court UK, and Google-sourced articles. They attributed errors to severe constraints: Alam's non-payment beyond nominal fees, tight deadlines, and his failure to engage promised UK counsel despite repeated advice. Samuel's July 10, 2025, witness statement asserted judgments were "read and arguments drawn," but verification lapsed due to resource limits, leading to "misapplications" without intent to mislead. MIO apologized, withdrew improper references, and argued errors were "inadvertent," not reckless, seeking dismissal of wasted costs and standard basis assessment elsewhere. They contended factual instructions came solely from Alam, positioning their role as structuring, not independent legal research.

Alam, initially aligning with MIO's withdrawal but later via new counsel Amr Bajamal of NHB Legal, accepted indemnity costs for the Defense but blamed MIO solely for defects, seeking wasted costs against them including refunded fees. He opposed indemnity for Transfer/Joinder, favoring standard basis, and denied Counterclaim liability, arguing Arabyads over-reviewed an unfiled document.

Legal Analysis

Justice Heath's reasoning rooted the decision in ADGM's adoption of English common law via the Application of English Law Regulations 2015, focusing on lawyers' duties as court officers. He affirmed the inherent jurisdiction for wasted costs, drawing from Myers v Elman [1940] AC 282 (HL), where Lord Wright described it as supervisory and compensatory for conduct defeating justice. The Privy Council in Harley v McDonald [2002] 1 NZLR 1 reinforced that only "serious dereliction"—beyond mere error, akin to "gross negligence"—triggers orders, exercised summarily without usurping regulatory roles.

Under ADGM-specific provisions, section 220 of the Courts Regulations 2015 mandates lawyers act "with independence in the interests of justice" and comply with conduct rules overriding inconsistent obligations. Rule 3 of the ADGM Court Rules of Conduct 2016 requires competent, diligent service; Rule 4 demands familiarity with laws, disclosure of relevant authorities (favorable or not), and avoidance of misleading statements. Breaches invite sanctions per Rule 10, including Rule 203 orders for "unreasonable or improper" conduct, mandating indemnity costs for wasted expenses.

The court dissected AI's role, citing empirical evidence like Magesh & Ors (2025) on LLM hallucinations—fabricated outputs hindering legal adoption. Justice Heath warned: lawyers must assume AI-revealed authorities "may not necessarily exist," placing verification onus squarely on them ( Ko v Li [2025] ONSC 2766). He referenced English cases R (Ayinde) v London Borough of Haringey [2025] EWHC 1040/1383 (Admin), where barristers faced wasted costs and referrals for unchecked AI, echoing Dame Victoria Sharp P's proviso: AI aids litigation but demands oversight to maintain justice confidence. Distinguishing, the court noted ADGM's flexible grounds (improper/negligent acts causing waste) exceed common law thresholds.

Applying facts, Heath rejected MIO's constraints as excuses—fees irrelevant without withdrawal under Rule 192—and found reckless AI use: superficial checks yielded non-existent cases (e.g., mis-spelled Johnson v Moreton [1980] AC 37 as 2014 High Court) and irrelevancies ( Re Lehman Brothers [2010] EWCA Civ 917 for insolvency, not ESOP rights). This breached Rules 4(4)-(6), misleading the court and wasting resources. Precedents like Skelmore Hospitality Group Ltd v Rosewood Hotel Abu Dhabi LLC [2019] ADGMCA 0001 deprecated disruptive conduct; Mingquo v Sadeghnia [2024] ADGMCFI 0005 at [90] supported sanctions. For quantum, Rule 199 favored Arabyads on indemnity, resolving doubts their way; standard basis per Rule 198 prioritized proportionality, reducing claims for overwork on unfiled matters ( Turcon v Assaf [2025] ADGMCFI 0002).

The analysis clarifies: AI hallucinations stem from human failure to verify, not tools—e.g., Google AI or BAILII misuse without reading full judgments. It distinguishes summary wasted costs from negligence suits, urging firms to implement verification protocols.

Key Observations

The judgment features pivotal excerpts emphasizing AI risks and duties:

• On verification imperative: "Lawyers using AI tools for research purposes should start from the premise that all authorities revealed by AI research may not necessarily exist... Without undertaking that verification task, the lawyer runs a serious risk that the Court may be misled." (Para 41)

• Echoing Ayinde [2025] EWHC 1383 (Admin): "Artificial intelligence is a powerful technology... but... carries with it risks as well as opportunities. Its use must take place... with an appropriate degree of oversight, and within a regulatory framework that ensures compliance with well-established professional and ethical standards." (Para 44, quoting Dame Victoria Sharp P)

• On MIO's conduct: "The issues that I have described... have all the hallmarks of the hallucinatory results that can occur when AI is used as a research tool... [It] was... reckless, and amounted to a breach of the Rules of Conduct." (Para 56(c), (e)-(f))

• Duties per Ko v Li [2025] ONSC 2766: "It is the lawyer's duty to ensure human review of materials prepared by non-human technology such as generative artificial intelligence... [and] not to mislead the court." (Para 48, quoting Myers J)

• Broader warning: "Reliance on technology does not dilute a lawyer’s core professional obligations." (Integrated from judgment principles, para 29-37)

These quotes, attributed to Justice Heath, highlight the court's balanced view: AI's legitimacy requires rigorous human oversight to prevent ethical breaches.

Court's Decision

Justice Heath ruled decisively on the costs applications, prioritizing compensation and deterrence:

1. MIO shall pay Arabyads AED 282,508 in wasted costs on indemnity basis for the Defense Application and related proceedings (AED 241,558 for review/hearings; AED 40,950 for the application itself). This covers unnecessary work reviewing the flawed Defense, assessed reasonably under Rule 199, with doubts favoring Arabyads.

2. Alam shall pay Arabyads AED 245,000 on standard basis for Transfer, Joinder, and Counterclaim Applications (reduced from AED 335,206.50 + AED 167,866.50 for disproportionality post-31 July 2025 on unfiled Counterclaim).

3. Defense Costs and Alam's Wasted Costs Applications dismissed; no order as to those costs.

No joint liability; MIO solely bears Defense fallout, shielding Alam. Implications are profound: reinforces lawyer accountability, potentially spurring AI training, policies, and tool enhancements (e.g., inaccuracy flags) per expert views in sources. Globally, it signals regulatory scrutiny—ADGM's ruling as "de facto regulatory signal" may inspire similar in UAE/elsewhere, affecting practice without public market ripples (private entities). Future cases could see stricter AI disclosures, reducing hallucinations' litigation waste while promoting ethical tech integration. For professionals, it's a mandate: verify or face penalties, ensuring justice's integrity amid AI's rise.