Article: Securing AI for Arbitration: Commercial and Government Advice
Steph Hunt
MCIArb; LLB (Hons); LLM (MIDS); Ciarb Australia Board Member; former adviser to Ministers for Foreign Affairs of the Commonwealth of Australia; Litigator, International Arbitrator and Mediator, and Sessional Academic at Monash University, Australia.
Assisted by Helin Usal, Thomas Hills and Sarah Khan (London City University) and the use of Legora AI
This paper follows the CIArb Guideline on the Use of AI in Arbitration 2025 and Summer Arbitration Retreat on Future of Work (Zurich) 2025
I. Introduction
Artificial Intelligence (“AI”) is permeating the world of arbitration. It is not only improving in its reliability at lightning speed, but one would struggle to find a young or tech-dynamic practitioner that is not using it in daily life from career-related tasks like salary negotiation or professional email writing to financial advice such as budgeting, travel itineraries, reflective support and health tips.
But they are not just leveraging AI for actionable guidance to improve various aspects of their daily lives, they’re also using it to complete aspects of their daily work. From Microsoft’s Copilot for email drafting and Adobe’s summarising features, to legal AI such as Legora, Harvey or Cocounsel, and even Chat GPT for (preferably anonymised, rather than specific) brainstorming, AI’s often unseen hand is reshaping our reality and redefining the future of work and efficiency.
Perhaps I come at this from a unique perspective. Since working for two former foreign ministers for two consecutive Australian governments, I increasingly view things through the prism of both private and public sector risk, with good public policy as an underlying and often over-riding consideration, alongside relevant security and data/ privacy risks. This leads me to look beyond the more obvious risk of error that can arise and appreciate the far-reaching security considerations when integrating AI into professional landscapes, including dispute resolution.
Arbitration, once perceived as a niche, less-costly and quicker forum for resolving private commercial disputes, as well as a strong neutral alternative forum as compared to state courts for foreign investor disputes, is now increasingly costly and lengthy due to its evolution and heightened sophistication. Time efficiency (and, by extension, cost efficiency) must remain a key priority in arbitration and can be partly achieved by leveraging AI. However, like all sectors, this cannot come at the cost of security and accuracy; balancing these considerations is critical. Since accuracy can be improved through robust professional guidelines and practices, this paper will focus on security.
In doing so, this paper will draw upon various principles, including those set out in the CIArb Guideline on the Use of AI in Arbitration (2025) (“CIArb Guideline“). Published in March 2025 as a hortatory instrument for the potential use of practitioners, primarily in commercial arbitration,[1] it provides arbitration practitioners and parties dependant on arbitration’s reliability with specific guidance in assessing the various specific legal AI use, in light of confidentiality and data risks that may arise. Consider it a resource for understanding more than just the simple fear of “AI errors.” I will explore hidden risks tied to data privacy, compliance challenges and national security implications, which are particularly sensitive in the context Investor-State Dispute Settlement (“ISDS”), where cross-border complexities and geopolitical concerns are especially relevant.
This analysis fundamentally reframes AI risks in arbitration, beyond the initial fear of AI hallucinations. Rather, practitioners should recognise a more pressing reality: AI requires rigorous data security consideration. As a sophisticated support mechanism, especially in its early stages of development, AI must be a tool that practitioners drive and control. By maintaining the reigns on quality control and strict attention to data security, arbitration practitioners and arbitrators can harness AI’s efficiencies without compromising the quality or confidentiality that defines so many arbitration proceedings, and without putting at risk any data around case strategy or other party material.
This paper encompasses a comprehensive framework for AI platform security assessment, jurisdictional risk analysis and practical implementation guidelines. It provides practitioners with the essential tools to navigate the transformative responsibly of working with AI, including through a step-by-step methodology for evaluating AI platforms, from encryption standards to data residency guarantees. It offers a systematic approach to due diligence that can be adapted across different practice areas and client requirements. Equally important, the jurisdiction-specific analysis reveals the stark differences in government access powers, from Sweden’s constitutional privacy protections to China’s mandatory cooperation requirements under the National Intelligence Law, highlighting how platform selection decisions carry profound implications for client confidentiality and professional obligations. Proactively confronting the issues addressed herein will help practitioners avoid reactivity and avoid the risk of compromise.
II. The Rise of AI Across Professions
I have always been interested in and, to the extent that I could, championed technological innovation as a cornerstone for societal advancement. From my campaign trail in Melbourne to my law firm’s business development, I have repeatedly witnessed AI’s transformative power. In terms of public uses, we see it in healthcare, where predictive algorithms diagnose diseases, and in education, where adaptive learning platforms personalise student experiences. Now, we see it deployed in law, where data-crunching and organisation tools revolutionise document review and automated processes which were previously undertaken by paralegals, interns and more junior practitioners and, crucially, subject to human error. While the introduction of powerful legal AI platforms may reduce the number of practitioners needed for tasks such as document review and production, it is equally likely to create new opportunities for those practitioners to oversee such tasks, or with future roles that do not even exist yet making aspects of legal services less costly.
The question of whether AI can or should be used in arbitration remains contested,[2] with recent legal challenges and professional sanctions demonstrating that acceptance is not yet universal.[3] The secondary question then becomes how it should be used and what level of client transparency, government oversight and safeguards should be mandated at policy and professional association levels. This issue is both regulatory and ethical in nature, concerning good professional practice and disciplinary matters that can occur with respect to professional negligence and failures of transparency in the arbitral process when disclosure is required.[4]
It appears that governments more generally are behind on providing citizens and businesses with programs for up-skilling to not only assist with leverage AI but also to upskill from jobs clearly at risk of extinction. Public policy, including as led by professional bodies, must consider educational programs and guidelines, much like the Chartered Institute of Arbitrators has done for the area of arbitration.[5] I am not supportive of over-regulation. But there seems to be somewhat of a vacuum in governments either using AI themselves or setting appropriate standards (which are not trade restrictive or prohibitive) to protect and support their citizens in their use of AI. This is evident by the lack of government white papers with real application in this space, but also in government activity in this space.
Beyond this obvious conundrum, while AI is commonplace in practitioners’ daily lives, let alone work lives, at the time of writing, there is virtually no thought leadership or academic input on the topic of security of using it in the legal sector, particularly international from an arbitration perspective, from reputable and longstanding authors, not to mention specific guidelines on how practitioners can assess various legal AI options. This reflection underscores a broader theme: the need for clarity about data integrity and security.[6]
This paper is intended to be read in conjunction with the CIArb Guideline which covers best practices for the use of AI in Arbitration[7], and provides draft clauses for tribunals to use to manage the use of AI, to set expectations in relation counsel’s use and the tribunal’s own use of AI, in the course the underlying arbitration. The CIArb Guideline is a stand-alone exception to the broader vacuum of thought leadership on the use of AI in Arbitration more broadly (not security focused)[8]. It also stands in contrast to more restrictive regulatory approaches, such as the New South Wales Supreme Court’s requirement for mandatory disclosure and court leave for certain AI applications,[9] and the Law Society of New South Wales’ emphasis on extensive risk management frameworks, as it outlines an adoptive approach based on consent of the parties.
III. Overview: AI in Arbitration
1. Common Tools and Functions
The use of AI seems uniquely suitable and applicable in the context of international arbitration than any other kind of domestic system, noting that the whole process of arbitration is underpinned and enlivened by consent of the parties. This means that it will be up to the parties to decide whether they wish to include its use in their arbitration, and this can be stipulated in the underlying trade agreement (for governments and foreign investors) or contract (for commercial parties), or adopted as a matter of procedure during an arbitration. Given the efficiency benefits, and AI’s ability to rapidly review and easily go through mountains of legal jurisprudence from various jurisdictions in multiple languages, it seems logical that many commercially-driven cross border parties may encourage the use of AI by both counsel and tribunals subject to appropriate governing clauses. This is likely why the CIArb Guideline favours an enabling approach while providing a high level risk-benefit analysis and draft clauses for tribunals to use to provide a framework for AI’s use in arbitrations, and why it does not stipulate any status quo requiring disclosure.[10] Given the sophisticated options for legal-AI that exist, commercial entities may conclude that it could even be against their interests, for their counsel, and even arbitrators, not to use AI.
Indeed, legal AI-driven solutions are already being widely adopted in arbitral processes for a variety of tasks that were formerly time-intensive and highly manual. These range from automated document review to the use of predictive analytics in case strategy. AI is particularly well suited to assisting counsel and arbitrators in international arbitration given the multitude of applicable laws and rules, whether substantive, procedural or arbitral seat rules, and the difficulty practitioners face in keeping abreast of these rules across multiple jurisdictions. Higher calibre legal AI platforms like Legora, Harvey, Libra, Cocounsel and DeepJudge offer significant advantages over general-purpose tools such as ChatGPT for professional legal work.
AI systems can rapidly process vast amounts of data across multiple languages, identifying patterns, inconsistencies and relevant information that would take human reviewers significantly longer. Law firms and arbitral tribunals are therefore experimenting with new workflows, seeking to streamline e-discovery processes and reduce the labour costs traditionally associated with sorting digital evidence. For instance, closed tasks like automated document categorisation and support with legal research tasks across multiple jurisdictions on sophisticated legal AI platforms offer more predictable cost savings and lower risk profiles compared to tasks such as strategic case analysis where AI outputs require more extensive human oversight and verification.
2. Considerations beyond efficiency
While the adoption of AI in arbitration brings tangible gains to both practitioners and parties saving substantial time, there is inherent risk in its use beyond the risk of error, for both commercial parties and governments. Many tribunals now view the introduction of AI as a competitive advantage, one that minimises procedural delays and enhances consistency across cases. Nonetheless, lingering concerns about data privacy, as addressed in Section IV, often surface in cases once the initial excitement of faster processes subsides.[11]
3. Early Warning Signals
Shifting Emphasis to Data Protection
According to the CIArb Guideline, arbitrators are not encouraged to delegate substantive decision-making to AI tools.[12] Instead, they may rely on AI much like they would rely on a tribunal secretary or a junior associate to summarise and research case materials. In such circumstances, overcoming the fact that internal workings of AI models are not easily explainable (commonly referred to as the “black box” problem), tends to be less of a concern. Because arbitrators maintain ultimate responsibility for arbitral awards, there is no requirement to audit every hidden layer of the AI’s reasoning. Expert judgement and legal analysis rest firmly in human hands, rendering the AI’s “black box” problem largely a non-issue for determining outcomes.
In any case, modern platforms such as Legora purposefully integrate traceable functionalities such as tabular searches which reference external legal authorities that are systematically flagged for user review. Rather than leaving an arbitrator to grapple with opaque reasoning, Legora provides source transparency so that practitioners can cross-check cited information, much as they would a physical library reference or research memo. For most applications especially in its early stages of development as experienced today, practitioners must view Legal AI as providing similar support to that traditionally provided by junior associates, paralegals or interns. In such applications, practitioners would not rely verbatim on the produced material but would use it as a research starting point, a basis from which to draft and would review and verify the output prior to any such reliance.
Given this dynamic, data protection emerges as the far more pressing concern. International arbitrations can involve sensitive commercial documents, proprietary corporate strategies and even sovereign data.
Layered security which covers encryption, sensible retention rules, and robust vendor checks, is vital in arbitration contexts. While “zero data retention” might sound ideal, it can be impractical because practitioners often need to revisit project records for audits or due diligence. Consequently, the question shifts from “How do we see into the AI’s black box?” to “How do we safeguard what the AI provider can see?” Platforms such as Legora, Libra, Harvey, DeepJudge, or even paid ChatGPT, generally cannot view or store the substance of users’ prompts, relying instead on encryption-in-transit, tagged backups, and role-based permissions. By balancing operational needs with strong security measures, arbitrators can leverage AI’s advantages without compromising sensitive data.
The more comprehensively secured and segmented the data environment, the lower the risk of inadvertent disclosure or malicious attack. Consequently, bolstering data safeguards on every level (device, platform, and infrastructure) is paramount, aligning with international best practices on handling confidential information in arbitral proceedings.
By approaching AI as a support mechanism rather than a decision-maker, and by maintaining strict attention to layers of data protection, arbitrators can harness the efficiencies of AI without succumbing to fears that an opaque algorithm might override human judgement or compromise privileged information.
IV. Beyond the ‘Error’ Narrative: Focus on Data Security and Privacy
1. Data Protection Regulations and Government Laws Underpinning Vendors
Public discourse often focuses on the risk of AI tools producing mistakes or “hallucinations,” yet the real challenge in arbitration is data security. Arbitration stands apart from litigation owing to its heightened confidentiality. Parties typically agree that no publication occurs, whether procedural or substantive, and this may only be changed by mutual consent. This privacy advantage, however, introduces new attack vectors for malicious actors, alongside regulatory and compliance obligations that vary widely across jurisdictions.
While the European Union (“EU”) General Data Protection Regulation (“GDPR”) imposes strict rules on handling personal data, it must be read together with the EU AI Act and other applicable data security legislation to understand the complete regulatory framework.[13] The EU AI Act now creates additional compliance obligations through its tiered penalty system, with fines up to €35 million or 7% of global turnover for prohibited practices[14]. In contrast, US state-level AI regulations impose significantly lower penalties, with Colorado’s AI Act capping civil penalties at US$20,000 per violation[15] and New York’s employment AI law fines range from US$500 to US$1,500[16]. This stark disparity in enforcement mechanisms reflects the broader philosophical divide; the EU’s comprehensive, rights-focused approach prioritising ethical considerations and user protections as compared to the US emphasis on innovation.
Moreover, arbitral disputes may not necessarily involve significant volumes of personal data, limiting GDPR’s direct application. In contrast to European nations which generally have checks and balances in this regard, Chinese AI providers may encounter heightened scrutiny under Article 7 of the PRC National Intelligence Law, which could allow authorities broad access to data, thereby conflicting with international standards, far beyond GDPR-related standards.[17]
Consequently, data protection in arbitration is less about meeting one specific legal regime (e.g. GDPR) and more about ensuring robust cloud security, encryption, and vendor trustworthiness. Parties therefore need to undertake an analysis on the provider, where the data is stored and on its encryption. This means they need to carefully map data flows, confirm that providers encrypt data in transit and at rest, and seek clarity on how (and where) data is stored and managed. When necessary, on-premises or localised solutions can further mitigate security risks, although this could involve storing a chip on-site at a higher cost and with diminished scalability. Currently, localised solutions rare and generally prohibitively expensive.
In practice, well-established AI providers often apply layered, cloud-based security frameworks that are regularly audited. Thus, any notion that GDPR compliance alone or any locally applicable privacy law suffices is misleading. The broader priority is protecting sensitive data, wherever it may be stored, through encryption, careful vendor vetting, and clear usage policies. By tackling vulnerabilities proactively rather than focusing narrowly on one regulation, arbitrators and counsel can preserve the high standards of privacy that define arbitration proceedings.
2. Further Government Access Risk Analysis
Government access to private data represents an under-considered yet critical risk when selecting AI platforms for international arbitration. The regulatory landscape varies dramatically across jurisdictions, creating fundamental differences that practitioners must understand when handling confidential matters where unauthorised government access could compromise legal professional privilege or expose sensitive commercial information to foreign state actors.
Sweden, for example, operates under constitutional privacy protections where authorities cannot access private data without judicial approval through formal court orders, with an extremely high threshold requiring demonstration of serious criminal activity or imminent national security threats. Swedish companies have constitutional rights to challenge government requests and cannot be compelled to build backdoors or compromise encryption.
The United States operates multiple legal authorities including the Foreign Intelligence Surveillance Act (“FISA”),[18] enabling US authorities to compel US companies to provide data stored anywhere globally, and National Security Letters (“NSLs”) allowing the FBI to compel data disclosure without court orders in national security investigations. However, crucially, US companies can legally resist government requests and have been largely successful in doing so, as demonstrated in the San Bernardino case where Apple successfully resisted FBI demands in 2016.[19]
China’s National Intelligence Law, Article 7, requires all organisations and citizens to “support, assist, and cooperate with national intelligence work,” creating a standing legal obligation with no right of corporate resistance or judicial challenge.[20] Chinese companies cannot refuse government data access requests or challenge them in court, representing the most permissive government access framework among major jurisdictions.
For practitioners handling confidential matters, these jurisdictional differences in government access powers represent a fundamental security consideration that must inform platform selection decisions. allowing government access with FISA Court approval at lower thresholds than traditional criminal warrants, the CLOUD Act (2018).[21]
For governments or counsel handling extremely sensitive and highly confidential data, an additional protection would be to host a chip on-site and ensure counsel and employees conduct localised AI processing.
V. Wider Relevance for Sovereign Stakeholders
The way arbitration professionals end up adopting AI can significantly impact stakeholders far beyond the arbitration community. Beyond sovereign stakeholders such as government agencies, investors and private companies of all sizes rely on consistent and enforceable mechanisms for resolving disputes. Ensuring that these mechanisms remain equitable, efficient, and secure is vital to preserving public trust in legal institutions and in arbitration.
Legal practitioners, arbitrators, mediators, policymakers, and technologists thus have a shared responsibility to create frameworks that safeguard data privacy and uphold the principles of fairness and due process in the context of using AI in arbitration. Continuing robust debate in think-tanks, bar associations, law societies, arbitration institutes and academic forums will help ensure that best practice guidelines remain current for use in arbitration and, where absolutely necessary, policy interventions addressing emergent and immediate AI challenges will assist in raising more awareness from legal practitioners and governments alike in this space.
The use of AI in ISDS presents particularly complex considerations that extend beyond technical capabilities to fundamental questions about appropriate use implementation readiness. Public interest and cross-border data regulations create a delicate operational environment where the threshold question of whether to adopt AI tools remains subject to legitimate professional debate. While potential efficiencies from AI applications, ranging from basic document review to complex legal analysis, may warrant consideration by some governments, the decision to proceed must account for significant implementation challenges including software acquisition costs, data security requirements, regulatory compliance across multiple jurisdictions, and the reality that many governments lack the technical, financial, or logistical resources for responsible AI deployment.
This requires practitioners to first determine whether AI use is appropriate for their specific context before conducting detailed analysis of platform vendors and document sensitivity classifications. Where governments conclude that AI adoption aligns with their capabilities and risk tolerance, bespoke arrangements may be necessary to leverage potential benefits while avoiding unjustifiable risks, potentially including local infrastructure hosting for the most sensitive applications where such options are technically and financially feasible.
VI. Conclusion: Charting the Future of AI-Enabled International Arbitration
The examination of leading AI platforms, from OpenAI’s enterprise-focused ChatGPT to emerging solutions such as Legora, illustrates both the diversity of available options and the critical importance of matching platform capabilities with specific security requirements for commercial and government parties. The analysis reveals that no single solution fits all scenarios. Rather, practitioners must develop sophisticated risk assessment capabilities that consider not only technical specifications but also the broader legal and regulatory environment in which they operate.
For ISDS, the stakes are particularly high. The intersection of national sovereignty, classified information, and cross-border data flows creates a uniquely complex environment where traditional risk assessment frameworks prove insufficient. The discussion of sovereign AI capabilities and on-premises solutions acknowledges that some governments may require complete data isolation for the most sensitive matters, even as the costs and technical complexity of such approaches remain prohibitive for most practitioners. But this should not hinder innovation. This reality underscores the need for calibrated, risk-stratified approaches that distinguish sensitive materials from highly confidential ones of national security significance.
Looking ahead, several critical developments will reshape the AI-arbitration landscape, particularly in ISDS. The emergence of sovereign AI capabilities reflects growing governmental awareness of data sovereignty issues, particularly as nations seek to maintain control over sensitive information processing. The evolution of hardware-level security solutions, including Intel SGX enclaves and dedicated processing chips, may eventually democratise access to high-security AI processing, though current cost barriers remain significant. Meanwhile, the inevitable standardisation of best practices across arbitral institutions will likely be driven by early adopters who establish robust governance frameworks that others will follow.
The broader implications extend well beyond the arbitration community itself. The choices made today regarding AI integration will determine whether these technologies become democratising forces that enhance access to high-quality legal services across jurisdictions, or create new barriers to justice through technological divides and security vulnerabilities. For developing nations and smaller legal markets, the risk of being excluded from an increasingly AI-driven arbitration ecosystem poses challenges to maintaining equitable dispute resolution mechanisms. This reality reinforces the vital importance of pro bono work in ISDS, work that I have been committed to since 2016 driven by the need to help put parties on an equal footing and my passion for ISDS given its strong public policy dimensions.[22] The emergence of AI tools now makes this commitment even more critical, as we must ensure that technological advances enhance rather than undermine access to justice. This demands proactive attention from international arbitration institutions and the broader legal community.
The analysis of government access risks reveals fundamental tensions between operational efficiency and data sovereignty that cannot be resolved through technology alone. While encryption and security enclaves provide important protections, the legal frameworks governing AI providers (including underlying security, privacy and data legislation, along with judicial review and company challenge powers) ultimately determine the extent of government access powers. This reality requires practitioners to develop new competencies in comparative law and international relations, understanding not just the technical specifications of AI platforms but also the geopolitical implications of their deployment.
The practical guidance provided in this paper for implementing AI in arbitration practice, from initial platform evaluation through ongoing security monitoring, reflects the reality that successful AI adoption requires sustained organisational commitment rather than one-time technology deployment. The emphasis on staff training, client communication, and continuous risk assessment acknowledges that AI integration is fundamentally a change management challenge that demands both technical competence and cultural adaptation within legal organisations.
The arbitration community faces fundamental questions about whether and how to integrate AI technologies into their dispute resolution practice. While some practitioners advocate for embracing AI’s potential efficiencies, others question whether such adoption is necessary or desirable given the associated costs, complexities and risks. The spectrum of potential AI applications, from basic drafting and document processing to sophisticated legal analysis, presents varying security and implementation challenges that demand careful evaluation. Where practitioners conclude that AI integration aligns with their specific context and capabilities, this requires comprehensive security frameworks, clear governance protocols and professional guidelines (e.g. the CIArb Guideline) about AI usage. However, the professional debate regarding appropriate AI adoption in arbitration practice remains in its early stages, with legitimate disagreement about optimal approaches across different types of disputes, client resources, and jurisdictional requirements. Any framework for AI integration must acknowledge this diversity, allowing for flexibility in its integration or lack thereof, based on party preferences while ensuring the technology enhances rather than undermines access to justice across jurisdictions.
The responsibility lies with today’s practitioners to ensure that the arbitration system of tomorrow remains worthy of the trust placed in it by parties. While some parties may think it would be unconscionable not to use AI to its full extent, others may embrace more selective applications of AI or resist integration entirely, opting for a more traditional approach.[23] Mutual understanding of the scope of disclosure duties on the use of AI is critical. In any case, security of the AI platform chosen and its uses, depending on the underlying context, is a critical consideration, and this framework provides practitioners with a roadmap for navigating AI integration in arbitration practice.
This does not ignore the broader and ongoing debate about preserving professional competence and the implications of potential impacts on junior legal positions and career development pathways within the profession. Arbitration is fundamentally based on party consent, and consent is found in the arbitration agreement driven by parties/ clients, whether commercial or government, and the process (once the arbitration is underway) is typically steered by arbitrators based on the relevant rules of the seat and influenced by parties. Therefore, where parties consent to the use of AI, they will retain discretion over AI integration decisions. In this vein, many will embrace comprehensive AI applications, while others will adopt selective uses (such as document classification systems for governments) and others will maintain more traditional approaches, each of which may represent the most responsible choice for their specific circumstances.
[1]Chartered Institute of Arbitrators, ‘Guideline on the Use of AI in Arbitration’ (March 2025) Preamble No.8 (discretion over use of AI by arbitrators), & Annex A (agreement on the use of AI in arbitration) <ciarb-guideline-on-the-use-of-ai-in-arbitration-2025-_final_march-2025.pdf > Accessed 22/07/2025.
[2] Recent cases demonstrate mounting professional sanctions for AI use in legal practice. In R (on the application of Frederick Ayinde) v. London Borough of Haringey [2024] EWHC 1040 (Admin), the High Court issued a £7,000 wasted costs order against counsel for citing five fabricated authorities, with the court noting: ‘On the balance of probabilities, I consider that it would have been negligent for this barrister, if she used AI and did not check it, to put that text into her pleading’ (para. 65). Similarly, in Williams v. Capital One Bank (D.D.C. Mar. 18, 2025), the court addressed ‘citation of nonexistent legal authority’ arising from AI-assisted pleadings. In Canada, Koh v Li (2025 ONSC 2766) resulted in sanctions where counsel submitted a factum referencing non-existent cases, with the court emphasising that ‘it is the lawyer’s duty to ensure human review of materials prepared by non-human technology such as generative artificial intelligence’ (para. 20). The Second Circuit’s decision in Park v. Kim, 91 F.4th 610, 613-626 (2d Cir. 2024), which referred an attorney for disciplinary action following admitted ChatGPT use, illustrates how AI-related professional liability extends beyond technical security concerns to fundamental questions of professional competence and client representation standards.
[3] The acceptance of AI in arbitration practice faces direct legal challenge in LaPaglia v. Valve Corporation (S.D. Cal. 2025), where parties are seeking to set aside an arbitral award allegedly because the sole arbitrator used ChatGPT to draft portions of the decision. The challenge argues that the arbitrator ‘betray[ed] the parties’ expectations of a well-reasoned decision rendered by a human arbitrator’ regardless of the award’s accuracy. Additionally, some counsel are now including express contractual stipulations prohibiting AI use in arbitrators’ terms of appointment, demonstrating proactive resistance to AI integration.
[4] UNCITRAL, ‘Model Law on Automated contracting’ (2024) <UNCITRAL Model Law on Automated Contracting (2024) | United Nations Commission On International Trade Law > Accessed 22 July 2025.
[5] Ibid.
[6] Ole Jensen, ‘Legal Tech in Arbitration: The EU AI Act’, 14th Baltic Arbitration Days 2025 (16 June 2025); Teresa Rodríguez de las Heras Ballell, ‘Use of AI by Arbitrators: CIAM-CIAR Principles’, 14th Baltic Arbitration Days 2025 (16 June 2025).
[7] Chartered Institute of Arbitrators, ‘Guideline of the Use of AI in Arbitration’ (N 1).
[8] Ibid.
[9] Supreme Court Practice Note SC Gen 23 (Practice Note) does not proscribe a general prohibition but does prohibit generative AI (Gen AI) being used to create documents for the purpose of providing evidence (including exhibits or affidavits) and an affidavit, witness statement or character reference must contain a disclosure that Gen AI was not used in generating the content – see paras 10-13. Expert reports similarly cannot use Gen AI to draft or prepare the content of an expert report (or any part of it) without prior leave of the Court – see paras 19-21. This only applies to cases in the Supreme Court of NSW.
[10] Chartered Institute of Arbitrators, ‘Guideline on the Use of AI in Arbitration’ (N 1).
[11] Chartered Institute of Arbitrators, ‘Guideline on the Use of AI in Arbitration’ Part IV (March 2025) <ciarb-guideline-on-the-use-of-ai-in-arbitration-2025-_final_march-2025.pdf > Accessed 22/07/2025. This part of the guideline addresses the use of AI in arbitration by arbitrators.
[12] Chartered Institute of Arbitrators, ‘Guideline on the Use of AI in Arbitration’ Part IV at point 8.2 (March 2025) <ciarb-guideline-on-the-use-of-ai-in-arbitration-2025-_final_march-2025.pdf > Accessed 22/07/2025. Arbitrators should not relinquish their decision-making powers to AI but may use AI to support more accurate and efficient processing of submitted information, always ensuring independent judgment. Arbitrators are advised to refrain from using AI in ways that could compromise the integrity of the proceedings.
[13] Regulation (EU) 2016/679 of the European Parliament and of the Council of 27 April 2016 on the protection of natural persons with regard to the processing of personal data and on the free movement of such data, and repealing Directive 95/46/EC (General Data Protection Regulation).
[14] Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonized 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) (Text with EEA relevance). Please see Chapter XII, Article 99 (3) for penalties.
[15] Colorado’s Artificial Intelligence Act (2024), Bill 24‑205 (2024), Concerning Consumer Protections in Interactions with Artificial Intelligence Systems (signed into law 17 May 2024; effective 1 February 2026) (codified at Part 17 of Article 1, Title 6 C.R.S., § 6‑1‑1701 et seq.).
[16] New York Workforce Stabilization Act, A 5429-A / S 1854-A, 2025–2026 Reg Sess (NY 2025) § 186-h.
[17] National Intelligence Law of the People’s Republic of China (as amended in 2018) Article 7. All organizations and citizens shall support, assist, and cooperate with national intelligence efforts in accordance with law.
[18] The Foreign Intelligence Surveillance Act (1978) (FISA) governs the surveillance and collection of foreign intelligence information within the United States. It establishes procedures for conducting electronic surveillance and physical searches for intelligence purposes. The act primarily targets foreign powers and their agents. The Foreign Intelligence Surveillance Court (FISC) oversees requests for surveillance warrants.
[19] In re Search of an Apple iPhone Seized During the Execution of a Search Warrant on a Black Lexus IS300, No. CM 16-10 (C.D. Cal. 2016).
[20] National Intelligence Law of the People’s Republic of China (as amended in 2018) Article 7 (N 13).
[21] Clarifying Lawful Overseas Use of Data Act (2018) allows the United States’ government to compel domestic technology companies to provide data stored on their servers, regardless of where teh data is located. The act also facilitates the creation of bilateral agreements with other countries and the United States to facilitate cross-border data access for law enforcement purposes.
[22] My pro bono involvement in ISDS began in 2016 with Trade Lab, headquartered in Geneva. My specialisation in this area was further developed through my government involvement as legal adviser to two Australian Foreign Ministers over two terms of government, which provided lived experience of ISDS from the state perspective. While setting up my firm in early 2023, I guest lectured in Trade Lab Clinics on ISDS at Monash University. Today, I continue this pro bono ISDS work independently, collaborating with counsel teams on ISDS cases, including to ensure equitable access to quality legal representation for developing nations and resource-constrained parties.
[23] LaPaglia v. Valve Corporation (S.D. Cal. 2025).