The Chartered Institute of Arbitrators has recently released its "2025 Guideline on the Use of AI in Arbitration" (the Guidelines). Split into four different sections, the Guidelines offer a useful framework for parties and arbitrators to incorporate and maximise the benefits of artificial intelligence (AI), including generative AI, in their arbitrations while mitigating certain obvious risks.

The Guidelines

Part I: Benefits and risks of the use of AI in arbitration

The Guidelines begin by setting out the various benefits of AI in an arbitration context, including improving efficiencies and cost-effectiveness in transcription, translation and the taking of evidence. The natural language processing capabilities of certain AI tools also lend themselves to carrying out research, analysing data and even generating text to be used in submissions or in the drafting of awards. This is consistent with the recent survey conducted by the Queen Mary School of International Arbitration, with 90% of survey respondents expecting to use AI for research, data analytics and document review in international arbitration, and 54% of respondents saying that their biggest driver for using AI was the expected time savings. 

As for the associated risks, one of the key concerns relates to the enforceability of awards which have used AI in the arbitral procedure. For example, if it was found that an arbitrator used an AI tool to assist in the decision-making process, perhaps in contravention of any agreement between the parties and the arbitrator as to the use of AI, then any subsequent award could be challenged on the grounds that due process had not been followed. In an English arbitration law context, this could fall under the ground for challenge in section 68(2)(c) of the Arbitration Act 1996 (failure by tribunal to conduct arbitration according to procedure agreed by parties). Furthermore, in considering the enforcement of any future award, parties will need to be mindful of the laws and institutional rules applicable to the arbitral proceedings, which may restrict the use of AI.

Another key risk inherent in the use of any third-party AI tool is confidentiality. Parties and arbitrators will need to be certain that the AI tool they are using is private and secure before inputting confidential data. Otherwise, the confidentiality of the arbitral proceedings could be jeopardised. Indeed, 47% of respondents to the Queen Mary survey identified confidentiality as a principal obstacle for the increased use of AI in arbitration.

The Guidelines also touch on the risk of algorithmic bias in certain AI tools, as well as the environmental impact of using generative AI, which often consumes significant energy and can lead to increased emissions.

Part II: General recommendations about the use of AI in arbitration

The take-home message of the Guidelines is that parties and arbitrators should make "reasonable enquiries" as to whether AI tools should be used in the arbitration, taking into account the benefits and risks outlined in Part I. In particular, parties and arbitrators should make reasonable enquiries as to any applicable AI-related rules, laws or regulations.

Perhaps most importantly, the Guidelines stipulate that (subject to any mandatory laws or agreement between the parties and arbitrators) any use of an AI tool during the arbitration shall not diminish the parties' and arbitrators' inherent responsibility and accountability in the context of the arbitral proceedings.

Part III: Parties' use of AI in arbitration

The key Guidelines in relation to the parties' use of AI are as follows:

• Upon receipt of a Request for Arbitration, arbitrators should review the arbitration agreement to consider what (if any) provision has been made for the use of AI. If the arbitration agreement is silent or unclear on this topic, the arbitrator should raise the issue of the potential use of AI at the appropriate time.
• Subject to applicable laws, the parties shall have autonomy to agree on the use of AI in the arbitration.
• Arbitrators may give directions or make procedural rulings with regards to the parties' use of AI, which could include:
  • appointing an "AI expert" to assist the arbitrator in understanding the AI tool in question;
  • requiring disclosure of the use of an AI tool by a party;
  • a ruling on the permissibility of the use of AI or a specific AI tool;
  • whether AI-generated or AI-assisted content should be admitted to the record; and
  • in deciding whether such content should be admitted to the record, requesting that parties make submissions as to how any inputs based on source data from the evidentiary record are linked to the machine-generated outputs of the AI tool.

It is envisaged that it would be most appropriate for arbitrators to exercise such procedural powers, for example: (a) where the use of AI might have an impact on the integrity of the arbitral proceedings; (b) to ensure the validity and enforceability of any ensuing awards; and (c) where there is a disagreement between the parties as to the use of AI. The Guidelines state that arbitrators should not regulate the "private use" of AI to the extent that this use does not interfere with the proceedings or the integrity of the arbitration.

Part IV: Use of AI by arbitrators

The key guidelines in relation to the arbitrators' use of AI are as follows:

• Arbitrators should not relinquish their decision-making power to AI.
• Rather, AI tools should be used to support more accurate and efficient processing of submitted information.
• Arbitrators must have special regard to the impact of the use of AI on the integrity of the proceedings and the enforceability of any future award.
• Arbitrators should independently verify any information obtained through AI.
• In any event, arbitrators must assume responsibility for all aspects of an award.
• Arbitrators are encouraged to be transparent about the use of AI and allow parties the opportunity to comment on and object to the use of any specific AI tool.

Comment

The Guidelines provide a useful practical resource for parties and arbitrators who may be considering using AI in their arbitrations. We may see the adoption of the Guidelines (or certain elements of them) into arbitral procedure going forwards (similar to the widespread adoption of the IBA Rules on the Taking of Evidence in International Arbitration). Indeed, the appendices to the Guidelines provide a model agreement between the parties on the use of AI in arbitration, as well as a model procedural order.

Of particular interest is the fact that the Guidelines appear to foresee parties (and perhaps arbitrators) using generative AI to analyse the evidentiary record and even to produce documents that will form part of such record. The Guidelines also envisage arbitrators using AI to "enhance…the quality of the arbitrators' decision-making". Use of generative AI in arbitral decision-making or in the evidentiary record is likely to raise issues of due process and procedural fairness, including the right to equality of arms and the right to properly challenge the opposing party's evidence. If the use of AI is adopted as a matter of course, this will likely lead to an increased number of challenges to arbitral procedure and/or the enforceability of awards as the arbitration community seeks to resolve the inherent tension between the desire for increased procedural efficiency and the integrity of the arbitral process.

As is apparent from the Guidelines, effective use of AI in arbitration requires significant cooperation and transparency between the arbitrator and the parties, as well as an arbitral tribunal that is sufficiently familiar with the use of AI. It therefore may not be appropriate to use AI in all arbitral proceedings, and parties and arbitrators should carefully weigh the benefits and risks before deploying AI tools in their proceedings.