- within Technology topic(s)
- with readers working within the Healthcare industries
- within Strategy, Tax and Family and Matrimonial topic(s)
A New Layer in Arbitral Practice
Artificial intelligence (“AI”) is rapidly becoming part of everyday legal practice, and arbitration is no exception. From drafting submissions to analyzing large volumes of documents, AI tools are increasingly present throughout arbitral proceedings.
The real question, however, is not whether AI is being used, but how its use should be managed within the arbitral process. In this respect, the need for regulatory and procedural guidance on the use of AI has increasingly come to the forefront. Various soft law instruments have already been introduced to address this emerging issue, and among them “Guideline on the Use of AI in Arbitration”1 published by the Chartered Institute of Arbitrators (“CIArb”) and the “SVAMC Guidelines on the Use of Artificial Intelligence in Arbitration”2 published by the Silicon Valley Arbitration & Mediation Center (“SVAMC”), reflect a clear shift: rather than restricting AI, the focus is on integrating it into the procedural framework in a controlled and transparent manner.
Where AI Fits into the Process
AI can appear at various stages of an arbitration, often in ways that are already becoming standard practice. Counsel may use AI tools to support legal research, draft submissions, summarize documents and analyze case law, particularly in disputes involving large volumes of data. AI is also increasingly used in document review and evidence analysis, including e-discovery, data classification and timeline construction.
In addition, AI can support hearing preparation and case management by assisting with the organization of hearing bundles, analyzing transcripts and comparing witness statements.
For arbitral tribunals, the use of AI is more limited but still highly relevant. AI tools may assist with document organization, reference checks or language refinement. This, however, raises a critical question: should AI replace the tribunal’s decision-making function? Under the current understanding and arbitral practice, the short answer to this question is no, AI’s role must remain strictly supportive. However, whether this position will remain unchanged, given the expanding scope of AI use and the speed of its growth, is far from certain.
Managing AI Through Procedure
Recent international guidance approaches the use of AI in arbitration through practical procedural tools. Of course, the use of AI can be examined across a wide range of issues, however, this article will focus on three recurring areas that emerge in how tribunals address AI within proceedings: disclosure, verification, and the limits of tribunal use.
- Transparency through disclosure
One of the preliminary questions is whether parties should disclose their use of AI. While not every use of AI requires disclosure, both the CIArb Guideline and the SVAMC Guidelines approach disclosure as a context-dependent procedural safeguard rather than a blanket obligation.
Under the SVAMC Guidelines, disclosure is not required as a general rule, but may become appropriate depending on the circumstances of the case, particularly where due process considerations arise. These include situations where AI tools materially influence submissions, evidence, or expert analysis, or where their use may otherwise affect the opposing party’s ability to understand and respond to the case. For example, this concern may arise where a party submits legal or technical analyses generated using artificial intelligence without verifying them, and the opposing party, unaware of that fact, is unable to challenge them effectively.
Similarly, the CIArb Guideline frames disclosure through the lens of procedural integrity and enforceability, noting that disclosure may be required where the use of AI has an impact on the evidence, the outcome of the arbitration, or otherwise affects duties owed within the proceedings.
The underlying rationale is, therefore, not transparency for its own sake, but the preservation of core arbitral principles:
- equality of arms,
- the right to be heard, and
- the integrity of the evidentiary record.
In practical terms, disclosure may extend beyond a simple statement that AI was used. As suggested in the SVAMC Guidelines, relevant details may include the tool used, its version, the manner of use, and, where necessary, the prompts and outputs relied upon.
Tribunals, in turn, are increasingly encouraged to address disclosure proactively through procedural orders. The CIArb Guideline explicitly recognizes the tribunal’s authority to impose disclosure obligations, define their scope and timing, and assess non-compliance, including through adverse inferences or cost consequences.
- Human oversight as a baseline
Human oversight is not merely a practical recommendation, but a foundational principle underpinning the use of AI in arbitration. Existing guidance consistently treats AI as a supportive tool outputs of which must remain subject to human review, validation, and responsibility.
Both the CIArb and SVAMC frameworks converge on a clear position: the use of AI does not reduce or redistribute responsibility within the arbitral process. Parties and their representatives remain under a duty to verify AI-generated outputs for factual and legal accuracy, with any errors ultimately attributable to the submitting party. More broadly, accountability remains fully human, regardless of the level of technological assistance.
This reflects a deeper procedural logic. AI may assist in generating or analysing information, but responsibility for what is presented as fact or law cannot be delegated. Human oversight therefore operates as a safeguard against risks such as hallucinations, unverifiable outputs and opaque reasoning processes.
At the same time, this baseline raises a forward-looking question. As AI systems become increasingly capable of cross-checking sources and performing complex validation tasks, the line between assistance and autonomous verification may begin to blur. Yet, the current regulatory approach remains firmly human-centric: even where AI can technically verify its own outputs, accountability cannot be transferred to the system.
For now, human oversight functions as both a practical necessity and a normative boundary, ensuring that professional judgment and procedural integrity remain anchored in human decision-making.
- Drawing the line for tribunals
The use of AI by arbitral tribunals raises more fundamental and structurally sensitive questions. While efficiency gains are evident, existing guidance draws relatively strict boundaries around permissible use.
Both the CIArb and SVAMC frameworks converge on a central principle: AI must not replace or interfere with the tribunal’s decision-making function. Arbitrators may not delegate any part of their mandate, particularly the assessment of facts, the interpretation of law, or the ultimate decision, and must retain full intellectual control over the award, regardless of any AI assistance.
This limitation is closely tied to core arbitral principles, including the duty to independently assess the case, the requirement to provide a reasoned award, and the need to minimise risks to the validity and enforceability of the award across jurisdictions.
At present, AI is therefore confined to a supportive role, typically limited to organizational and technical functions such as document management, reference checks or language refinement. However, this boundary is not purely technical; it is normative.
Looking Ahead
The more difficult question is not where the line for use of AI in arbitral proceedings currently lies, but whether that line is stable.
As AI systems become more accurate, explainable and context-aware, it is not inconceivable that their role in arbitration will inevitably expand and AI may begin to perform functions resembling legal reasoning. From the perspective of tribunals’ role, this raises a provocative question: could AI one day participate in, or even lead, parts of the decision-making process, with human involvement reduced to a final layer of review?
Current guidance does not permit such a shift and remains firmly grounded in a human-control model. However, as AI systems become increasingly sophisticated, the possibility of more autonomous models, even in areas approaching decision-making, can no longer be dismissed outright.
Whether arbitration will ever encounter fully automated systems remains uncertain. What is clear, however, is that the focus must shift toward designing and maintaining effective balance mechanisms. In that sense, the challenge is not simply to accommodate AI, but to continuously recalibrate the limits of its use in a way that preserves the integrity of the process.
A related issue is the consequence of failing to disclose the use of AI where disclosure is warranted, or of using AI beyond the permitted scope set by the tribunal or the applicable procedural framework. In such cases, the consequences are likely to be procedural rather than technological: tribunals may draw adverse inferences, give reduced evidential weight to affected submissions or evidence, allocate costs against the non-compliant party, or take such conduct into account when assessing procedural fairness and the integrity of the record. In more serious cases, particularly where undisclosed or improper AI use affects due process or the reliability of the award, such conduct may also create enforceability risks at the post-award stage.
In that sense, the boundary is not fixed, but dynamic.
Click here to read this article, along with all other newsletter articles, on our newly launched, reader-friendly microsite.
Footnotes
1. Chartered Institute of Arbitrators (Ciarb), Guideline on the Use of AI in Arbitration (2025), available at: https://www.ciarb.org/media/bpndtcgu/guideline-on-the-use-of-ai-in-arbitration_updated-sept-2025.pdf
2. Silicon Valley Arbitration & Mediation Center (SVAMC), SVAMC Guidelines on the Use of Artificial Intelligence in Arbitration (1st Edition, 2024), available at: https://svamc.org/wp-content/uploads/SVAMC-AI-Guidelines-First-Edition.pdf
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
