Consideration of Artificial Intelligence in Arbitration Terms of Reference

As artificial intelligence (AI) continues to evolve and integrate into various aspects of legal practice, counsel and arbitral tribunals drawing up their Terms of Reference (TOR) establishing the terms of the dispute being referred to arbitration and also formulating their procedural orders should consider the implications of AI. This client alert highlights the importance of addressing AI in TOR negotiations and provides an overview of likely topics international arbitration practitioners can expect to treat in TORs and procedural orders.

Use Cases for AI in Arbitration

AI can be utilized in many ways to enhance the efficiency and effectiveness of arbitration proceedings, including, but not limited to:

• Document review: AI can manage and review large volumes of documents quickly, making it easier to identify relevant evidence or responsive documents, reducing the time and cost associated with document review.
• Legal research: AI-powered legal research tools can help arbitration practitioners quickly identify relevant case law and precedents for further review, saving time and potentially improving the quality of legal arguments.
• Contract review: Some AI platforms have been specially trained to analyze contracts. With these tools, arbitrators can focus on contract provisions that govern the issues at hand, even if that text is scattered throughout the contract.
• Summarize: AI can summarize documents, including transcripts and other evidence, with varying levels of detail.
• Predictive analytics: AI can forecast trends based on historical data. AI algorithms can analyze historical cases and predict likely outcomes or settlements based on various factors, helping parties make more informed decisions about pursuing arbitration or negotiation.

Guidelines for AI Use in Arbitration

The Silicon Valley Arbitration and Mediation Center (SVAMC) and more recently the Chartered Institute of Arbitrators have published guidelines covering AI applications, consent and disclosure norms, safeguarding confidentiality, and maintaining diligence in AI use. The rapid development of AI is constantly presenting new challenges, including concerns related to AI biases, hallucinations, privacy concerns, and due process issues. Comprehensive and enforceable AI use guidelines would assist to ensure the ethical and responsible use of AI in arbitration.

Arbitration Terms of Reference Likely Will Need to Include AI Clauses

The Terms of Reference (TOR) is a creature of international arbitration that identifies the names and contact information of the parties and arbitrators, the claims and relief being sought, the key issues to be decided, applicable laws and procedural rules, and place of arbitration. In essence, it is a contract executed by the parties to an international arbitration and the arbitrators that establishes the terms of the dispute being referred to arbitration. In some cases, a TOR can be interpreted to supersede the parties’ underlying arbitration agreement, especially if there are differences between the two. Together with the arbitrators’ procedural order, the TOR contains the parties’ and arbitrators’ agreement on the rules of engagement for the arbitration proceeding.

As of this writing, it is unlikely that most parties or arbitrators are including AI clauses in their TORs or procedural orders at the outset of an arbitration. That is likely to change. We anticipate that consideration of agreements on the parameters for AI use will soon become a standard topic for TOR negotiations as well as negotiations for procedural orders.

Summarized below are some of the topics that any TOR or procedural order with an AI clause is likely to address:

Scope of AI Use

TOR clauses, or, failing that, arbitrators’ procedural orders, should establish the permissible uses of AI by parties to an arbitration and the arbitrators. AI is likely to be used, to some extent, by parties and arbitrators for document review, legal research, and quasi-legal arbitration management functions. Whether parties or arbitrators may use AI for drafting assistance may be a point of friction in negotiations as the use of AI for substantive legal writing is not universally agreed upon.

Transparency and Disclosure

If the disclosure requirements in the SVAMC guidelines and many U.S. judges’ standing orders reflect the current norm, it is more likely than not that parties and the tribunal will want to include provisions concerning the disclosure of the use of any AI tools in a TOR, or, failing that, in their procedural order. Requirements for disclosure of AI use may vary based on the AI tool used and its impact on an arbitration. Additionally, the extent of disclosure is likely to be a subject of debate, as parties may disagree about whether simply disclosing AI use is enough or whether parties need also to identify specifically parts of any documents drafted by AI, what AI program or programs(s) were utilized (including their functionalities, limitations, and potential biases), and whether and to what extent any AI-generated content was verified by a human. The disclosure requirements surrounding how AI is used may be a particularly fraught topic in relation to the topic of electronically stored information (ESI) protocols or other aspects of the document disclosure process, given that prompts given to AI to assist in refining the identification and culling of responsive documents may contain mental impressions, conclusions or opinions of counsel that is protected from disclosure by applicable privileges.

Human Oversight and Decision-Making

AI use clause negotiations are also likely to include a discussion about whether AI tools are allowed to replace human judgment. For arbitrator use in particular, comparison of the delegation of their decision-making role to AI with the use of arbitral secretaries is instructive. The International Chamber of Commerce Note to Parties and Arbitral Tribunals on the Conduct of the Arbitration under the ICC Rules of Arbitration (January 2021) provides a useful reference point. This proposes that a tribunal could seek the assistance of a tribunal secretary to research matters or even draft correspondence, procedural directions or even the award itself. But that “[u]nder no circumstances may the arbitral tribunal delegate its decision-making functions to an administrative secretary or rely on an administrative secretary to perform on its behalf any of the essential duties of an arbitrator.”

Cases where parties have challenged an arbitral tribunal’s delegation of drafting responsibilities to a secretary have reached courts. The theme which has emerged is deference from the courts to how arbitrators use tribunal secretaries to administratively assist them, and even draft their awards, as long as parties had been provided an opportunity to comment on the use of a tribunal secretary in advance. Similarly, it is likely that proper consideration of AI tools up front, and inclusion of their use and scope within a TOR, could insulate tribunals from later challenges that they have improperly delegated their deliberative functions.

Relying on AI to assist research and drafting has another potential pitfall: the right to be heard is fundamental to international arbitration. Courts have agreed previously that arbitrators have failed in their duties when their awards have decided a matter based on legal theories or external facts which had not been previously raised with the parties. Thus consideration in the TOR of the scope of independent use of AI for research by arbitrators versus reliance on information from the parties could reduce the potential for later debate over the proper arbitral procedure.

Confidentiality and Data Protection

The protection of any matter-sensitive and confidential data should also be at the forefront of any AI clause negotiations. It is highly likely that parties will want the confidentiality of any data input into AI tools to be maintained, and to mandate that any AI use comply with relevant data and privacy protection laws.

Inclusion of protocols for assuring confidentiality in the use of AI up front in a TOR may be helpful to avoid any potential disputes later on if it emerged that party representatives or arbitrators used public AI systems or even inadvertently provided those systems confidential facts about the matter.

Objections

Parties and arbitrators may also have concerns about procedures for counsel to raise objections to the use of specific AI tools or for specific AI use cases. Any AI clause should therefore address the grounds for counsel to raise objections to the use of AI and the procedure for the arbitral tribunal to decide those objections.

Costs

Lastly, any AI-related costs should be addressed along with the allocation of other arbitration costs. The parties will need to establish if and how AI use will be allocated as expenses amongst the parties.

Set Aside and Emerging Regulatory Risks

As noted above, it is possible that improper delegation of decision-making functions to AI or using AI without proper disclosure and verification could lead to challenges of arbitral decisions or awards. Any such challenges are likely to be based on due process violations or non-compliance with a TOR or other procedural requirements of the forum.

Relatedly, while AI use regulation is evolving rapidly, the example of the EU Artificial Intelligence Act (Regulation (EU) 2024/1689) could apply to arbitrators established in the Union when its near-full implementation is enforced (currently due for 2 August 2026). That regulation would require “to the extent the deployer exercises control over the input data, that deployer shall ensure that input data is relevant and sufficiently representative in view of the intended purpose of the high-risk AI system.” A “high-risk” system is defined to include systems “intended to be used by a judicial authority or on their behalf to assist a judicial authority in researching and interpreting facts and the law and in applying the law to a concrete set of facts, or to be used in a similar way in alternative dispute resolution.”

Key Takeaways

Although arbitrations to date have not typically included AI-related clauses in TORs and procedural orders, we anticipate that inclusion of AI-related clauses in these documents will soon become commonplace. Parties to arbitration and arbitrators alike must become knowledgeable about AI and understand its benefits, limitations, and risks.

For further information or assistance, please contact our team at Crowell & Moring LLP.