Practical Realities Of AI In International IP Arbitration

Introduction

We are in an age where artificial intelligence-enabled legal applications have flooded the legal world overnight. International and domestic arbitration-specific AI tools compound the number of applications useful to general commercial disputes practices.¹ Layer on top AI programs aimed exclusively at intellectual property, and IP arbitration practitioners and arbitrators are awash in AI.² What’s more, international arbitrators must also navigate the emergence of AI laws and guidance from countries, arbitral institutions and other alternative dispute resolution (ADR) organisations, committees and working groups. In this chapter, we consider the practical implications of these developments.

The chapter first discusses the definitions of AI that are relevant to the IP arbitration community, which are many and growing, and examines how applicable laws, institutional standards and party‑ or tribunal‑imposed rules interact across jurisdictions. It then reviews the current landscape of AI‑related standards and guidance issued by arbitral institutions and ADR bodies, before addressing the growing use of AI predictive analytics and the professional and procedural issues this raises.

The chapter proceeds to highlight several instructive cases that have already confronted AI‑related issues, and it concludes by briefly considering future developments, including the potential transformation of hearings through real‑time AI use.

The definition of AI and why it is important

Defining artificial intelligence is a natural starting point. The numerous definitions – no singular definition is widely accepted³ – range from the scientific to the philosophical.⁴

For international IP arbitration, the relevant definitions of AI are those provided by the following:

• applicable governing law;
• governing arbitral institutional rules;
• agreed upon party-supplied or arbitrator-imposed protective orders; and
• the law governing enforcement actions.

This disparate list of overlapping sources gives rise to significant variation in what constitutes AI. For example, in Regulation (EU) 2024/1689 on artificial intelligence, the European Union defines “AI system” as a:

machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments.⁵

The Regulation also sets out that the definition of “AI system” should revolve around key characteristics of artificial intelligence, such as “its learning, reasoning or modelling capabilities, distinguishing it from simpler software systems and programming approaches”.⁶ That definition predominantly, although not exclusively, revolves around generative AI, a branch of AI that generates original, human‑like content (such as text or images) based on machine‑learning models trained on large datasets.⁷

The Silicon Valley Arbitration & Mediation Center (SVAMC), which is dedicated to global technology dispute resolution,⁸ defines AI as “computer systems that perform tasks commonly associated with human cognition, such as understanding natural language, recognizing complex semantic patterns, and generating human-like outputs”.⁹ This SVAMC definition subsumes various types of AI.¹⁰

As yet another example, the England & Wales Courts and Tribunals Judiciary’s Artificial Intelligence Guidance for Judicial Office Holders defines AI as “computer systems able to perform tasks normally requiring human intelligence”¹¹ and goes on to separately define subsets of AI, such as generative AI, machine learning and technology-assisted review.¹² The Guidance, which is updated annually, refers to AI tools generally, but breaks out certain types of AI to detail specific concerns. While the general definition of AI has remained unchanged, additional definitions and explanations have been added to reflect the constantly changing AI environment.¹³

These examples are just three of the many definitions across a myriad of jurisdictions. However, many arbitral institutions, courts and countries are also continuing to create different definitions of AI, alongside rules, regulations and guidelines for its use.¹⁴Not surprisingly, these entities tend towards broad, all‑encompassing, potentially future‑proof, definitions of AI using variations of the above-mentioned definitions.

The breadth of variation creates a need to ensure those definitions impose clear, understandable boundaries on those participating in international arbitration. Many of the definitions leave significant gaps. The parties in IP disputes must therefore take particular notice of the definitional options to ensure there is zero exposure not only to confidentiality and data protection concerns, but also to tools that have less than robust capabilities to assess the complexities inherent to IP disputes, which could result in the need to counter misinformation that, in the IP space, once tangled can prove difficult to untangle.

Definitional concerns could also create avenues for award invalidation or raise enforcement obstacles. For example, at what point does an arbitrator’s use of AI cede the exercise of sufficient independent judgement? This scenario creates interesting arguments in jurisdictions such as France, where only natural persons may act as an arbitrator.¹⁵ In the absence of consideration of the potentially applicable definitions, allowing or prohibiting AI use could affirmatively violate relevant law or could constitute a waiver of potential challenges.

The rapid advances in AI use and continuously developing laws pose an acute concern where arbitrations may take years, during which the definitional and regulatory landscape may change significantly.

Evolving AI standards, guidelines and rules

Of course, elements of AI have been used in arbitration for many years, largely in the context of document review and disclosure. Arbitrators have readily endorsed the use of machine-led review and predictive coding tools to minimise the time and cost of document review. Recently, however, generative AI tools have also become widely available to parties and arbitrators. In response, arbitral institutions and ADR organisations have adopted differing approaches to AI use, many of which have been articulated in guidance issued in the past year. Many of the institutions are creating AI standards with the stated aim of drawing IP and technology disputes to their institutions,¹⁶ including by espousing broad principles to provide guardrails for party and arbitrator conduct. However, those principles leave significant discretion to the parties and arbitrators in how they use AI. The following examples are worth particular attention.

In 2025, the Chartered Institute of Arbitrators (Ciarb) released its Guideline on the Use of AI in Arbitration. Ciarb’s guidelines detail the numerous novel issues confronting AI use, including bias, due process and the “black box problem”.¹⁷ The tension between imposing outright limitations on AI use and the transparent use of AI by the parties permeates each of the guidelines discussed in this chapter and others, making it a key distinction among how the various guides approach AI use. Ciarb’s guidelines focus on disclosure more than restrictions,¹⁸ but do empower the arbitrator to dictate the extent of AI use in a proceeding, subject to certain limits.¹⁹ Ciarb’s guidelines rightly draw specific attention to disclosure of AI use, not only by expert witnesses but also by fact witnesses.²⁰ Expert witness use of AI is already a well-known concern,²¹ while fact witness use has garnered less attention. The emergence of AI provides a mechanism for fact witnesses to more easily familiarise themselves with the issues at stake, which has the potential to influence their testimony.

The International Chamber of Commerce (ICC) published an “AI governance and standards” policy paper in July 2025.²² The ICC’s approach stems from its recognition that there is a fragmented set of international standards and that a one-size-fits-all approach will not account for the different sizes and resources of the parties arbitrating before it.²³ The policy paper advocates for the development of a global, harmonised approach that is driven by the commercial market.²⁴

The American Arbitration Association-International Centre for Dispute Resolution released its AAAi Standards for AI in ADR, which “provide a comprehensive framework for responsible AI use in ADR”.²⁵ Each of the six standards are separated into guidelines for administrators, neutrals and advocates, reflecting the differing considerations applicable to each group. The administrator standards are unique compared to those of other institutions, which typically do not discuss specific standards for administrators. Those standards recognise – and serve as a critical reminder – that administering organisations’ systems, many of which increasingly rely on in-house AI tools,²⁶ should be taken into account when selecting an arbitral institution. For IP arbitration practitioners (and their clients) in particular, this may require data protection to factor into forum selection in contracts.

The China International Economic and Trade Arbitration Commission (CIETAC) released its Provisional Guidelines on the Use of Artificial Intelligence Technology in Arbitration.²⁷ Although CIETAC expressly states that the guidelines are not part of its arbitration rules, they do appear to be rule-oriented (eg, using the term “shall” as well as declarative requirements). One important takeaway from CIETAC’s guidelines is the recognition that national AI laws and regulations may impose specific restrictions, while much may be left to the discretion of the tribunal in setting the parameters for AI use. For example, section 3.1 allows parties to agree on whether and how to use AI “[u]nless otherwise provided by law or otherwise decided by the arbitral tribunal”.²⁸ Section 5.1 similarly suggests parties include AI use parameters in arbitration clauses “[u]nless otherwise provided by law”. With evolving national laws and regulations – not only in the arbitral forum but also in the area of enforcement – this requires practitioners and arbitrators to vigilantly monitor and assess changes in the AI laws of multiple jurisdictions.

The Ministry of Justice of the People’s Republic of China and the China National Intellectual Property Administration issued the Guiding Opinions on Strengthening Arbitration of Intellectual Property Disputes at the end of 2025.²⁹ The guidelines expressly aim to “enhance the arbitration of intellectual property disputes” and develop “specialized rules” for IP arbitration.³⁰ Further, the guidelines call for the creation of a list of national arbitration institutions that have a goal of increasing their IP qualifications. The guidelines repeatedly recommend collaboration with companies to develop IP-specific rules and experience, and are notable because of the potential impact on companies doing business in China that use or contribute to standard essential patents. The guidelines implicitly indicate that China appreciates the need to offer a trusted enforcement mechanism and dispense with questions of arbitrability that may have historically dissuaded IP enforcement in China.

More IP-specific rules and guidelines have been issued by SVAMC and JAMS. SVAMC was an early adopter of guidelines on AI use in arbitration.³¹ Its guidelines are notable for three reasons. First, SVAMC focuses on global technology dispute resolution, making it one of the most directly applicable and tailored sets of AI guidelines for IP arbitration. Second, the guidelines were prepared by a committee of experienced practitioners and arbitrators from the global arbitration community, imbuing the guide with a cross-section of practical experience. Third, SVAMC opened the draft for public comment, allowing it to consider a broad spectrum of views before issuing its final guidelines.

As a result, the principles espoused in the SVAMC guidelines reflect specific considerations. These range from identifying AI tool “policies on recording, storage, and use of prompt or output histories” to heightened disclosure obligations. While SVAMC does not advocate for mandatory disclosure of AI use, it states that certain circumstances may warrant disclosure of sufficient information, including the complete prompt used and associated output, to “help reproduce and evaluate the output” of the AI tool.³² That level of disclosure certainly raises work product doctrine issues, but its applicability to expert opinions, for example, could prove invaluable – if not entirely warranted³³ – given the black box problem associated with AI-generated output.³⁴ In one of the more specific caveats on an arbitrator’s use of AI, and one not inconsistent with the black box concern, SVAMC cautions arbitrators about AI’s ability to generate outside-the-record information and advises them that they must independently verify citations and any uncited statements.³⁵

JAMS is one of the few arbitral institutions to issue rules expressly applicable where “the disputes or claims are AI-related”.³⁶ It has adapted its Comprehensive Arbitration Rules and Procedures to create the Artificial Intelligence Disputes Clause and Rules.³⁷ These Rules require JAMS to propose arbitrators capable of “evaluating disputes involving technical subject matter with appropriate background and experience”. In addition, the Rules contain a provision restricting expert discovery of “AI systems and related materials” to a “secured environment”. US patent litigators will no doubt find this provision far less detailed than the protective orders issued by US courts in patent infringement cases, which often contain highly specific, multi-page provisions governing, for example, the inspection of source code and prototypes. Nevertheless, it is a key acknowledgment. The Rules also permit arbitrators, upon the parties’ request, to appoint a third-party expert to inspect the AI system at issue and give evidence on the inspection results.³⁸ Although it seems highly unlikely that either party would relinquish expert analysis to a third party in an IP arbitration, the rule suggests that JAMS has given forethought to the potentially expansive scope of “AI-related” disputes or claims.

Almost all companies – whether acting voluntarily, incentivised by market forces or required to do so by government regulation – require internal and third-party data breach prevention policies. While parties have no choice but to accept whatever security systems that courts put in place, parties do have a choice and responsibility to ensure that their chosen ADR forum’s systems are insulated. This concern becomes more important as ADR institutions rapidly develop and implement AI-enabled systems.³⁹

Each of the above rules and guidelines makes one thing very clear: practitioners need to incorporate these principles into explicit provisions on the use (or limitations on the use) of AI during disputes. Practitioners may also want to convince their colleagues and clients who are drafting arbitration provisions in contracts to seriously consider including provisions expressly addressing AI use. This is particularly true in IP cases, where the arbitration provisions may receive greater attention than run-of-the-mill commercial contracts.

Use of AI as the “arbitrator” raises critical considerations

Cynics may declare that AI can never replace the independent exercise of an arbitrator’s judgement. Certainly, legitimate concerns exist regarding “AI arbitrators”, including AI systems’ lack of empathy towards parties,⁴⁰ the potential for exponential replication of inherent bias in decision-making⁴¹ and uncertainty over whether AI can create new applications of law.⁴² Also, while unlikely to be an issue in an IP arbitration, AI cannot rule ex aequo et bono (ie, render a ruling based on fairness rather than a strict application of the law).⁴³

Despite this, AI arbitrators are already being deployed. None are (yet) capable of handling (or are trusted to handle) complex commercial arbitrations, let alone arbitrations with the complexity of IP disputes. But AI’s case-predictive abilities are already being used and, with the rapid expansion of AI’s capabilities, are likely to become standard practice.⁴⁴

AI-driven predictive analytics⁴⁵ use prior judicial and arbitral decisions to assess the facts fed into the AI and, as the name suggests, predict expected outcomes.⁴⁶ These predictive abilities extend to the ultimate chance of success on the merits of specific claims, potential damages and how a particular judge or arbitrator – based on the history of their prior decisions – could influence that chance of success, as well as any damages expectation.⁴⁷ These predictive tools are already in use by, among others, litigation funders to determine whether certain cases are wise investments.⁴⁸

The implications of predictive analytic use by practitioners are both positive and negative, depending on which “side of the aisle” you sit and how your adversary employs it. For example, using predictive analytics to select arbitrators could lead to tribunal selection as a true battleground. Parties have long selected arbitrators based on reputation and predilections, but AI predictive analytics takes this to a whole new, highly detailed level. In trade secret disputes, parties can identify arbitrators who take expansive or narrow views of what constitutes a trade secret. In patent disputes, parties can identify whether arbitrators are more likely to invalidate patents, find infringement or award significant damages. This predictive capability raises many questions, such as whether arbitral institutions need to impose restrictions on AI use for this purpose or their own use of AI in choosing their rosters or selecting the non-party-appointed chair to “balance the scales”. Additionally, the use of this technology could create colourable grounds to object to an arbitrator’s appointment by using arguments about their track record.

Another example is whether practitioners should or must use predictive analytics in formulating legal strategies and specific arguments. Rules of professional conduct may implicitly mandate it. Alternatively, concerns over malpractice claims may warrant it. This may not be merely an issue of practitioners insulating themselves from such claims, but also one that ensures compliance with malpractice insurance policies.

By using predictive analytics, practitioners can identify strengths and weaknesses in the case, and play to the strengths and more effectively address the weaknesses. In patent cases, this may implicate, among other issues, claim constructions and prior art combinations. Trademark practitioners (and experts) could create predictive customer confusion studies and parse trademark decisions across jurisdictions to compile extensive lists of decisions supporting their arguments and negating counterarguments.

For arbitrators, this raises a host of considerations. Should arbitrators use – or be permitted to use by arbitral institutions or parties – predictive analysis at all? The risk of undue bias in “knowing” who should win and why at the outset of a matter is paramount, particularly where equitable claims are brought. Similarly, relying – explicitly or implicitly – on arguments, case authority or even evidence that the parties did not address raises issues over the decision-making process and how “reasoned” the official award may be.

Practitioners need to determine whether to include permissions or restrictions on predictive AI use in contractual arbitration provisions and procedures governing the arbitration, or whether they can be included in procedural rules or orders. Arbitrators need to determine (and should be knowledgeable enough to assess) whether AI tools should be allowed or blocked in IP arbitrations, in particular. That knowledge is critical where the parties disagree on AI use, thereby requiring the arbitrator or arbitrators to exercise independent judgement. They should also consider whether use of AI could create a risk of award invalidation based on assertions of inherent bias, lack of due process or the black box problem. In addition, parties and arbitrators need to be aware of invalidation or enforcement risk based on claims that the arbitrator exceeded their authority or failed to use independent judgement due to the use of AI. Arbitral institutions must similarly determine whether to directly address this with a definitive position in their rules or expressly require arbitrators to take a position on it on a case-by-case basis.

Ultimately, predictive analytic capability could result in “a counter-trend” favouring “human-centric justice” to the exclusion of “AI-based justice”.⁴⁹

Public decisions shed light on these concerns

Courts are already grappling with issues concerning the use of AI in arbitration.

The LaPaglia v Valve Corporation case marks the first test of whether AI-assisted drafting can undermine the validity of an arbitral award. The respondent received an AAA award in its favour on the claimant’s antitrust and unfair competition claims. The claimant petitioned a US district court to vacate the award on the basis that the arbitrator exceeded his authority. Specifically, the claimant alleged that the sole arbitrator “outsourced his adjudicative role” based on a “ChatGPT authorship test” that indicated AI-generated language in the award.⁵⁰ The claimant also alleged that the arbitrator “mentioned the use of ChatGPT for writing” the award and his “desire to complete proceedings” before planned travel. The respondent moved to dismiss the petition, which the court granted based on lack of federal court jurisdiction. The court did not address (even in dicta) the claimant’s allegations that the arbitrator abused the use of AI in drafting the award. As a result, the case does not expressly provide any guidance regarding the use of AI in arbitration, but it does show that its use requires careful consideration. The claimant refiled his petition to vacate the award in California state court re-alleging, among other faults, the arbitrator “outsourc[ing] his adjudicative award to Artificial Intelligence”.⁵¹ Because that petition remains pending, it is an open question whether the California state court will rule or otherwise comment on the claimant’s AI-misuse allegations.

In the United Kingdom, a court awarded wasted costs where one party cited hallucinated cases in briefing and further referred the counsel to the Bar Standards Board and the Solicitors Regulation Authority.⁵² Although the professional regulators did not initiate contempt proceedings, they emphasised the duty of lawyers to comply with their professional obligations. The Bar Council of England and Wales, for example, has published guidance on using ChatGPT and similar generative AI tools, cautioning that generative AI’s ability to “inadvertently generate information disorder, including misinformation, is a serious issue of which to be aware”.⁵³ Similar cases in US courts are now legion and have extended to questioning whether expert reports using AI are subject to rejection.⁵⁴ Arbitrators are certainly encountering the same issues and assessing how fees and costs should be awarded in such cases.

In addition, US courts are ruling on how the use of AI impacts the US attorney–client privilege and work product doctrine, which, put most simply, allow parties to withhold from discovery communications with counsel and materials reflecting an attorney’s impressions or analysis of a case.⁵⁵ For example, in United States v Heppner, the defendant, anticipating indictment, used a generative AI tool to prepare a defence strategy.⁵⁶ After retaining counsel and following indictment, the defendant asserted that the resulting documents were protected by the attorney–client privilege and the work product doctrine.

The court rejected both arguments based on classic applications of the law on those protections (ie, the AI-generated documents were not privileged because they were not communications between the defendant and his counsel and did not constitute attorney work product because they did not reflect the counsel’s views and were not prepared at the counsel’s direction). The defendant had not retained counsel at the time the documents were created and the work product doctrine did not apply retroactively under the circumstances of the case. This case highlights the risks associated with both lawyer and fact witness use of AI systems. Whether and how circumstances such as the use of open or closed systems (eg, publicly accessible versus in-house systems where data does not feed into a public AI database) may cause waiver or trigger exceptions to the work product doctrine is beyond the scope of this chapter. Nevertheless, they are interesting issues to ponder where fact and expert witnesses and practitioners are routinely using AI.

Conclusion and outlook

The future of AI use in arbitration will be constrained only by the rules, regulations and practices agreed to by the parties or imposed by the arbitrators. These will have to change and be reimagined repeatedly as AI capability continues to expand exponentially.

As noted above, AI is already being employed – albeit subject to human oversight – to handle Uniform Domain-Name Dispute-Resolution Policy, construction and maritime disputes. Aside from these instances, arbitral institutions’ use of unredacted awards to train AI could rapidly expand its use in arbitral decision-making.

The use of AI during hearings could radically reshape opening and closing arguments and witness testimony. For example, cross-examination and redirect examination could be altered in real time by feeding in testimony as it is elicited and having the AI process it against the database of case evidence to identify lines of questioning. This could allow parties to generate demonstratives during the hearing, for example, to help depict and contextualise evidence.⁵⁷ It could further enable arbitrators to identify inconsistencies or to understand complex issues in near real time, thereby promoting greater understanding or, where applicable, enabling them to pose questions during the hearing that may otherwise be noticed only after the fact. AI’s translation abilities will also expedite proceedings and bridge language barriers for non-native speakers of the language of the arbitration.⁵⁸

While we are optimistic that greater use of AI can improve the effectiveness of arbitration, there are also likely to be negative consequences. In addition to the concerns noted above, accuracy of the outputs, confidentiality and data security risks, lack of transparency (eg, the black box problem) and over-reliance on AI technology are all live risks. Together, these could well lead to increases in inaccurate submissions and help generate frivolous filings. These negative consequences are likely to be of heightened concern in the context of arbitration because arbitral proceedings and awards largely remain confidential, which may act as a barrier to arbitrators learning from one another’s encounters with AI.⁵⁹

Regardless of any personal views on AI, for current practitioners the adoption of AI in arbitration is likely to be the most radical change to practice we will see in our working lives.

Acknowledgement

The authors would like to thank Alexandria Moriarty and Anisha Sandhu for their invaluable insights and contributions, and Kara Battista for her unrelenting support of the authors.

Footnotes

1 On 4 March 2026, the American Arbitration Association announced the launch of its “Resolution Simulator”, a case-predictive outcome model that forms part of its AI Arbitrator tool. The sheer number of legal‑focused AI tools is astounding and constantly growing. Accordingly, the identification of available applications is beyond the scope of this chapter and, regardless, would quickly grow stale. Therefore, this chapter generally avoids naming specific tools.

2 The majority of IP-focused AI tools are non-litigation applications (eg, patent searching, analytics and drafting; automated market monitoring; and clearance); however, programs targeted at or applicable to IP disputes do exist and are expanding rapidly. Any AI chatbot would be happy to generate a list of those tools. For a comprehensive explanation of the kinds of available AI applications, see Elizabeth Chan, Kiran Nasir Gore and Eliza Jiang, “Harnessing Artificial Intelligence in International Arbitration Practice”, 16(2) Contemp. Asia Arb. J. 263, 267-73 (November 2023), https://ssrn.com/abstract=4648246.

3 See Guidelines on the Use of Artificial Intelligence in Arbitration, Silicon Valley Arbitration & Mediation Center (SVAMC), (2024), at 13, https://svamc.org/wp-content /uploads/SVAMC-AI-Guidelines-First-Edition.pdf; see also Gizem Halis Kasap, “Can Artificial Intelligence (‘AI’) Replace Human Arbitrators? Technological Concerns and Legal Implications”, 2(2) J. Disp. Resol., 211 (2021), https://scholarship.law.missouri.edu/jdr/ vol2021/iss2/5. Kasap’s article offers a succinct, easily digestible definitional understanding of AI. See id., at 211–15.

4 See Peter Stone, et al., “Artificial Intelligence and Life in 2030: One Hundred Year Study on Artificial Intelligence: Report of the 2015 Study Panel” (September 2016), at 12–17, http://ai100.stanford.edu/2016-report.

5 Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence and amending Regulations (EC) No 300/2008, (EU) No. 167/2013, (EU) No. 168/2013, (EU) 2018/858, (EU) 2018/1139 and (EU) 2019/2144 and Directives 2014/90/EU, (EU) 2016/797 and (EU) 2020/1828 (Artificial Intelligence Act).

6 ibid.

7 Guideline on the Use of AI in Arbitration (2025), Chartered Institute of Arbitrators (Ciarb), https://www.ciarb.org/media/bpndtcgu/guideline-on-the-use-of-ai-in-arbitration_ updated-sept-2025.pdf.

8 See SVAMC, footnote 3.

9 See id., at 8.

10 See ibid.; see also Chan et al., footnote 2, at 267–68 (explaining types of AI).

11 See Artificial Intelligence Guidance for Judicial Office Holders, Courts and Tribunals Judiciary (31 October 2025), at 2, https://www.judiciary.uk/wp-content/uploads/2025/10/ Artificial-Intelligence-AI-Guidance-for-Judicial-Office-Holders-2.pdf.

12 id., at 2–3.

13 For example, adding definitions for “algorithm”, “alignment” and “responsible AI”, among others, and expressly stating that “judges must always read the underlying documents”. See ibid.

14 “Given the potential for various national laws to apply – for instance, an arbitration seated in Paris, governed by Mexican law, with hearings in Hong Kong – it becomes necessary to harmonise the potentially disparate local and international standards relating to the use of AI.” SVAMC, footnote 3, at 13.

15 See Kasap, footnote 3, at 238, note 201 (delineating the approach in different jurisdictions); see also Crenguta Leaua and Corina Tănase, “Artificial Intelligence and Arbitration: Some Considerations on the Eve of a Global Regulation”, Revista Română de Arbitraj, Vol. 17, Issue 4 (December 2023), at p. 38.

16 See Aaron Wininger, “CNIPA and MOJ Issue Guiding Opinions on Strengthening Arbitration Work in Intellectual Property Disputes”, China IP Law Update (24 December 2025), https://www.chinaiplawupdate.com/2025/12/cnipa-and-moj-issue-guiding-opinions-on-strengthening-arbitration-work-in-intellectual-property-disputes/; “JAMS Announces New Artificial Intelligence Disputes Clause and Rules”, JAMS (23 April 2024), https://www.jamsadr.com/news/2024/jams-announces-new-artificial-intelligence-disputes-clause-and-rules (rules apply to AI-related disputes).

17 Ciarb, footnote 7. These issues are addressed in detail by others, including, Kasap, footnote 3, and Allyson Reynolds and Paula Melendez, “AI arbitrator selection tools and diversity on arbitral panels”, International Bar Association, https://www.ibanet.org/ article/97cb79fa-39e9-48c1-8cb0-45569e2e62af. The “black box problem” describes a user’s inability to understand exactly how an AI tool arrived at the generated output. See SVAMC, footnote 3, at 15 (commentary to Guideline 1). For a comprehensive analysis of this issue, see Kasap, footnote 3, at 229–32.

18 Compare id., at section 4.4, with sections 7.1 to 7.7. Section 4.5 restricts the arbitrator’s ability to limit “private AI use” unless that use would “interfere with the proceedings and the integrity of the arbitration process” or would otherwise not be in line with “the domestic courts in the relevant jurisdiction”. id., at section 4.5.

19 Compare id., at sections 4.1, 4.3 and 4.7, with section 5.1.

20 See id. at section 7.3.

21 See David Ball and Alexandria Moriarty, “The irony – using generative AI in a case about the dangers of generative AI”, Reuters, (30 January 2025), https://www.reuters.com/legal/ legalindustry/irony-using-generative-ai-case-about-dangers-generative-ai-2025-01-30/; see also Kohls v Ellison, No. 24-CV-3754, 2025 WL 66514, at *3–5 (D. Minn. 10 January 2025) (reprimanding counsel for submitting an expert report containing hallucinations); Concord Music Grp., Inc. v Anthropic PBC, No. 24-CV-03811-EKL, 2025 WL 1482734, at *3 (N.D. Cal. 23 May 2025) (calling the issue of hallucinated citation(s) in expert report “a serious one – if not quite so grave as it first appeared” and striking one paragraph in the expert report); Matter of Weber, 85 Misc. 3d 727, 741–43 (Sur. Ct. 10 October 2024) (detailing problems with the expert’s use of AI to perform calculations, which could not be reproduced by the judge’s own prompt in the same AI tool; “[t]he mere fact that artificial intelligence has played a role, which continues to expand in our everyday lives, does not make the results generated by artificial intelligence admissible in court.”).

22 “AI governance and standards”, ICC (July 2025), https://iccwbo.org/wp-conten t/uploads/sites/3/2025/07/2025-ICC-Policy-Paper-AI-governance-and-standards.pdf.

23 See “Artificial intelligence”, ICC, https://iccwbo.org/global-insights/artificial-intelligence.

24 See “AI governance and standards”, footnote 22, at 8–9.

25 AAAi Standards for AI in ADR, American Arbitration Association-International Centre for Dispute Resolution, https://www.adr.org/media/sx2mjcdj/aaai_standards_for_ai_in_ adr_.pdf. The Vienna International Arbitration Centre Note on the Use of Artificial Intelligence in Arbitration Proceedings similarly suggests high-level principles (see https://www.viac.eu/wp-content/uploads/2025/04/VIAC-Note-on-AI-1.pdf). The Stockholm Chamber of Commerce Arbitration Institute’s “Guide to the use of artificial intelligence in cases administered under the SCC rules” offers high-level caution on the use of AI (see https://sccarbitrationinstitute.se/wp-content/uploads/2025/01/scc_guide_to_ the_use_of_artificial_intelligence_in_cases_administered_under_the_scc_rules-1.pdf).

26 See, eg, AI Arbitrator, AAA, https://www.adr.org/ai-arbitrator/.

27 Guidelines of the China International Economic and Trade Arbitration Commission on the Use of Artificial Intelligence Technology in Arbitration (Trial), CIETAC (18 July 2025), https://www.cietac.org/articles/33749 (translated with machine translation).

28 ibid. Section 5.2 suggests tribunals “invite[]” party opinion on the use of AI, which highlights the risk, particularly in IP disputes, that tribunals may be unwilling to impose particular restrictions on AI use.

29 See Wininger, footnote 16.

30 ibid.

31 See SVAMC, footnote 3.

32 See id., at 10.

33 See, eg, footnote 21.

34 See footnote 17.

35 SVAMC, footnote 3, at 12 (Guideline 7).

36 JAMS Artificial Intelligence Disputes Clause and Rules (14 June 2024), https://www.jamsadr.com/artificial-intelligence-disputes-clause-and-rules.

37 ibid.

38 Ciarb’s guidelines contain a provision allowing arbitrators to appoint AI experts to assist the arbitrators in understanding whether certain tools should be used. See Ciarb, footnote 7, at section 4.2.

39 All the guidelines generally address confidentiality and data security concerns and suggest parties and arbitrators review how the AI tools store and use information that is input or uploaded to the tools. Ensuring confidentiality and data security due to technology changes long precedes the proliferation of AI tools.

40 See Cole Dorsey, “Hypothetical AI Arbitrators: A Deficiency in Empathy and Intuitive Decision-Making”, 13 Arb. L. Rev. (2021); Kasap, footnote 3, at 232–36; Dr Aline Tanielian Fadel, “Predictive Analytics and Diversity in International Arbitration: Friends or Foes?”, Am. Rev. of Int’l Arb. (8 October 2021), https://aria.law.columbia.edu/predictive-analytics-and-diversity-in-international-arbitration-friends-or-foes/.

41 See Reynolds and Melendez, footnote 17; Kasap, footnote 3, at 242–45.

42 See Kasap, footnote 3, at 249–50.

43 See Chan et al., footnote 2, at 293, note 69.

44 See Kathleen Paisley and Edna Sussman, “Artificial Intelligence Challenges and Opportunities for International Arbitration”, NYSBA New York Dispute Resolution Lawyer, Vol. 11, No. 1, at 39 (Spring 2018); SVAMC, footnote 3, at 12.

45 Predictive analytics differ from predictive coding or technology-aided review, in which AI performs the review for potentially relevant documents and those responsive to discovery requests.

46 See Chan et al., footnote 2, at 272; see also Maxi Scherer, “Artificial Intelligence and Legal Decisionmaking: The Wide Open?”, 36 J. of Int’l Arb, 539, 546–54; Ciarb, footnote 7, at 4.

47 See Scherer, footnote 46, at 546–54 (explaining studies in which AI predicted US Supreme Court and European Court of Human Rights decisions with surprising accuracy); see also Kasap, footnote 3, at 215–21 (describing and commenting on the same studies).

48 See Ana Fernández Araluce, “AI in International Arbitration: Unveiling the Layers of Promise and Peril”, Iurgium (previously Spain Arb. Rev.) (2024), 35, 37; see also Myriam Gicquello, “Artificial Intelligence in International Arbitration”, The Oxford Handbook of International Arbitration (Oxford University Press, 2020), 600–01; Verity Jackson-Grant, “Will artificial intelligence revolutionise the litigation funding market any time soon? (Spoiler alert) Probably not!”, Practical Law Dispute Resolution Blog (27 September 2019), https://uk.practicallaw.thomsonreuters.com/w-047-1972?transitionType=Default& contextData=(sc.Default)&firstPage=true.

49 See Leaua and Tănase, footnote 15, at 43.

50 See LaPaglia v Valve Corp., No. 3:25-cv-00833-RBM-DDL, 2025 WL 3527053, at *2 (S.D. Cal. 9 December 2025).

51 See LaPaglia v Valve Corp., No. 25CU068896C, at 1 (Sup. Ct. Cal. 26 December 2025).

52 Frederick Ayinde v The London Borough of Haringey [2025] EWHC 1040 (Admin), https://www.judiciary.uk/wp-content/uploads/2025/05/Ayinde-v-LB-Haringey-Judgment-Ritchie-J-03.04.25-HD-2.pdf.

53 “Considerations when using ChatGPT and generative artificial intelligence software based on large language models”, The Bar Council, at 5 (30 January 2024), https:/www.barcouncilethics.co.uk/wp-content/uploads/2024/01/Considerations-when-using-ChatGPT-and-generative-artificial-intelligence-Nov-2025.pdf.

54 See generally, footnote 21.

55 See, eg, Concord Music Grp., Inc. v Anthropic PBC, No. 24-CV-03811-EKL, 2025 WL 1482734, at *2 (N.D. Cal. 23 May 2025) (finding prompts and outputs are work product protected and citing Tremblay v OpenAI, Inc., No. 23-cv-03223-AMO, 2024 WL 3748003, at *2–*3 (N.D. Cal. 8 August 2024)).

56 See United States v Heppner, No. 25 Cr. 503, 2026 WL 436479, at *2 (S.D.N.Y. 2026).

57 See Chan et al., footnote 2, at 289.

58 ibid.

59 See Kasap, footnote 3, at 221–36 (detailing lack of training data and numerous other issues).

Originally published by GAR.