Keeping up with the machines: how arbitral institutions are responding to AI (Part 1)

The LCIA

The LCIA's current Rules were published in October 2020, several years before the proliferation of advanced large language model-based AI (LLMs). AI is consequently absent in the Rules, although the LCIA's ongoing rules revision consultation specifically identifies the impact of AI on arbitration as a focal theme. It is therefore possible that rules around disclosure, safeguards, due process, evidence integrity and enforceability may be introduced into the LCIA rules in the future. Whilst no official launch date for the new rules has been shared, a draft version is expected for community feedback in late 2026.

The ICC

The most recent ICC Rules came into effect in June 2026, and AI was a conspicuous omission. This absence is, in itself, surprising in the face of the increasing popularity of AI – the 2024 Report on the State of the UK Legal Market by Thomson Reuters found that 91% of respondents either use or predict that they will use AI at least for conducting document review. Further, the 2026 Rules do not shy away from codifying the use of technology elsewhere. By way of example, Rule 3.1 requires parties to make written communications with the Secretariat by electronic communication as default and Rule 38.1(a) permits electronic signatures on awards.

However, it is possible that the ICC has decided that a fast-developing area such as AI is better dealt with by practitioner guidance and case-specific tribunal directions. The removal of the list of case management techniques from the Rules, and their replacement with more flexible Secretariat guidance notes, means that guidance can be updated comparatively quickly. It is therefore possible that the ICC will use these guidance notes as a vehicle for AI governance instead of more unwieldy rules. Indeed, the existence of the Task Force on Artificial Intelligence in Dispute Resolution suggests that practical guidance on arbitration is a real possibility.

The SIAC

The SIAC's most recent Rules were published in 2025 and also lack any express provisions on AI. However, the SIAC publishes Practice Notes on procedural matters, including on cases administered under the SIAC Rules, and in an interview with LexisNexis, the SIAC's Registrar, Vivekananda Neelakantan, stated that the SIAC "remains open to support the responsible and informed adoption of AI and technology in arbitration". It is therefore possible that the adoption of AI could come through a Practice Note.

The 2025 Rules include Rule 61, which encourages parties and tribunals to "agree on reasonable measures to protect the information that is shared, stored, or processed in relation to the arbitration". In the LexisNexis interview, Neelakantan stated that the "underlying objective of this provision is to underscore the growing importance of data protection in light of the increasing use of digital tools in modern arbitration proceedings". If arbitrating in the SIAC, parties should be aware of this increased focus on data protection when planning what AI tools they use, and how.

UNCITRAL Rules

The UNCITRAL Rules similarly do not yet include any explicit provisions on AI. However, UNCITRAL has explored how AI could be utilised in arbitration through a questionnaire for States and a Colloquium on the use of AI held in February 2026. A recurring theme in the discussion is maintaining the integrity of the arbitration process and retaining cognitive independence and impartiality. In particular, it sounds a note of caution about delegating reasoning to AI and the need to ensure protection from algorithmic thinking distortions and cultural bias. If the UNCITRAL Rules are updated, these themes could well be incorporated.

Arbitration, by its nature, is flexible, and parties tend to have a broad degree of discretion as to the procedural framework. Even if there are no specific provisions in respect of AI within the rules of an institution, this should not preclude the use of AI in the course of the arbitration, not least because parties and tribunals can utilise existing rules as a vehicle for the use of AI.

Efficient and expeditious proceedings

The LCIA's Rule 14.1(ii), the ICC's Rule 23.1, the SIAC's Rule 3.5 and UNCITRAL's Article 17 all provide that proceedings should be efficient and expeditious. If the use of AI could speed up the arbitration process, then its use should be facilitated by these clauses. Further, the LCIA's Rule 14.5 states that technology can be employed "to enhance the efficiency and expeditious conduct of the arbitration (including any hearing)". Similarly, the ICC and SIAC also specify that proceedings should be cost-effective: the time saved on potentially burdensome tasks by using AI could significantly reduce costs.

Parties should be aware, however, that there may not be parity in terms of the quality of the AI tools utilised. For a party with access to frontier AI models or purpose-built legal AI, significant time and costs could be saved. However, if parties have access only to free public AI models this will significantly undermine how much time they could save and the accuracy of the AI. Parties without adequate training in prompt construction may receive inaccurate or unhelpful responses, and it is possible that they will have to spend more time reviewing and correcting the AI than they would have spent doing the task manually. Further, parties using public AI models risk breaching confidentiality rules and / or data protection regulations. Therefore, parties (and, where relevant, tribunals) will need to consider whether the use of AI is in fact cost-effective and appropriate. In addition, if parties have agreed upon an environmentally friendly arbitration, they should consider whether the use of AI aligns with this aim.

Order of the tribunal

The LCIA's Rule 14.5, the ICC's Rule 23.2, the SIAC's Rule 32.2, and UNCITRAL's Article 17.1 give the relevant tribunal power to conduct the arbitration and make orders as it deems appropriate. Accordingly, if the use of AI fundamentally aligns with the goal of efficient and expeditious proceedings, and it is appropriate for the tribunal to order its use, it could do so.

The parties' agreement

Under the LCIA's Rule 14.4, the ICC's Rule 20 and the SIAC's Rule 3.6, parties can agree on joint proposals for the conduct of arbitration. This might include leveraging AI in some or all of the following ways (proposed in the UNCITRAL Questionnaire):

• Selecting arbitrators or mediators: AI could be used to identify suitable arbitrators, to flag conflicts of interest and / or check disclosures by potential arbitrators.
• Drafting legal documents: AI could be used to draft legal documents including arbitration agreements and clauses, submissions and the tribunal's decisions.
• Quantifying damages: AI could be used to analyse financial data, apply valuation models, and streamline calculations, particularly in complex disputes.
• Case management: AI could be used to automate scheduling, send reminders to the parties and manage deadlines. It could also be used to provide transcription and translation services.
• Decision making: At the most extreme end of the spectrum, AI could be used fully to decide less complex cases.

Whilst these uses could result in considerable time and cost savings, retaining the integrity of the arbitration is key; if this is undermined, the award could be set aside. Under the Arbitration Act  1996 (AA 1996), s 68(2)(i), any "irregularity in the conduct of the proceedings" is a ground for a party to apply to set aside the arbitral award. By way of example, if a document on which an award was based was drafted (in part or in whole) by AI and found to be inaccurate, this irregularity could result in the award being unenforceable.

Before parties consider using AI in arbitration, they should first ensure that there are no laws in the arbitral seat that could conflict with its use. This is particularly pertinent to data protection laws, where restrictions could be placed on how parties process data.

To the extent that parties intend to utilise AI bilaterally in the course of an arbitration (as opposed to unilaterally for discrete tasks behind the scenes), they should agree from an early stage how it is to be used. If parties instead wait until a dispute arises about the use of AI, this is likely to be costly and labour-intensive to resolve. Further, if no records of AI usage have been kept, it may be difficult to delineate AI work product from human work product which, in turn, could effectively leave all documents and evidence open to challenge.

Parties should therefore agree:

• how AI tools, and the use of AI tools, will be defined in the arbitration;
• what types of AI tools parties can, and cannot, use;
• when and how the use of AI should be disclosed to other parties and the tribunal, including whether any AI generated documents should be watermarked to reflect their origin; and
• if and how records should be kept of usage.

Some arbitral rules, such as the SIAC's Rule 3.5, require tribunals to endeavour to ensure the enforceability of the award. Tribunals may order parties to agree on how AI will be incorporated into the arbitration at an early stage if doing so will decrease the likelihood that an award will be unenforceable due to AI misuse. The tribunal should create a procedural order that reflects the parties' agreement – this could be bespoke or follow existing templates, for instance the template procedural order found in Appendix B of the Chartered Institute of Arbitrators' (CIArb) Guideline on the Use of AI in Arbitration (2025).

Tribunals should be cautious if they themselves use AI. If a tribunal chooses to summarise submissions or evidence using AI, it should take care that the AI summaries are accurate and do not miss any key details. Further, whilst tribunals may use AI for operational efficiencies (such as organising or summarising data), they must clearly delineate between that and their analytical decision-making processes. It is already established, under the rules of the arbitral institutions (including the LCIA), that tribunals should not delegate their decision making. In cases such as P v Q & Ors [2017] EWHC 194 (Comm), delegation of decision making by the tribunal was found to be grounds for the removal of the tribunal members under section 24(1)(d)(i) of the AA 1996. The use of AI by tribunals poses a similar risk: in Association des ressources intermédiaires d'hébergement du Québec (ARIHQ) v Santé Québec, the Québec Superior Court annulled an arbitral award after finding that the arbitrator had effectively delegated the drafting of reasons to a third party by basing their decision on hallucinated AI cases.

In the US, in LaPaglia v Valve Corporation [No. 3:25-cv-00833 (S.D. Cal. Apr. 8, 2025)], a Californian claimant contested an adverse arbitral decision on the grounds that the arbitrator relied on AI to such an extent that he "outsourced his adjudicative role." Although the court declined to rule on the merits of the AI-related allegations (on the basis that it did not need to do so; the court dismissed the case due to a lack of subject-matter jurisdiction), this is a clear demonstration that the use of AI can and has been challenged in relation to arbitral awards.

Although some arbitral institutes have been slow off the mark in the AI race, there are others that have embraced it and incorporated it into their rules. In the second part of this article series, we will examine three of those institutions and how their AI frameworks operate in practice.