AI and International Arbitration – Watching Brief: AI in legal research and arbitrator selection

This is the next instalment in our series of Watching Briefs examining the intersection of artificial intelligence and international arbitration practice. In this instalment, Joshua Kelly and April Lacson focus on two of the most immediately practical applications for arbitration practitioners: legal and arbitrator research.

The use of AI in the legal industry is increasing substantially. A 2026 survey noted that 92% of legal professionals use at least one AI tool as part of their daily work. Another study found that 78% of law firms surveyed either already treat Generative AI (GenAI) as central to their workflow or expect to do so in one to two years. The same study found that 74% use AI specifically for legal research.   

For arbitration practitioners, the attraction to AI is obvious: legal professionals devote an average of 19% of working hours to legal research. The development of GenAI and agentic AI, a form of artificial intelligence that can act on and develop workflows and tasks autonomously, can radically reduce the time spent on this task. 

That said, as discussed below, AI tools vary in quality and their use must be consistent with a lawyer’s ethical and professional responsibilities and the risks they present. 

Overview of AI tools for legal analysis and arbitrator research

AI tools for legal and arbitrator research range from general-purpose large language models (LLMs)-based GenAI systems to law-specific and arbitration-specific AI tools. These tools can be used instead of traditional legal research and restrictive searches using Boolean operators by allowing prompts based on ordinary language and generating creative user-defined outputs such as comparative tables, mind maps, slides and memos: 

• General-purpose LLM-based GenAI tools are widely accessible, often offering free or subscription-based options. Depending on the AI tool, they may also offer agentic AI capabilities. However, while they can rapidly synthesise large volumes of information, their outputs are drawn from largely unvetted and undifferentiated sources. They are also untrained on the specificities of legal work and language. Using them for legal and arbitrator research will therefore require a greater degree of scrutiny than law- and arbitration-specific AI tools. 
• Law-specific AI tools and “wrapper” services built on general-purpose AI models draw on curated legal databases and are trained on legal datasets. Some of these AI tools also enable secure uploads of confidential client materials, which can be integrated with case or arbitrator research to generate context-sensitive legal analysis.  
• Other legal research platforms offer arbitration-specific AI tools. These are trained on arbitration datasets and are marketed to be sensitive to the nuances of arbitral practice and research. They provide access to arbitration-specific, expert-verified sources. These AI tools also work with arbitrator databases and analytics tools, allowing AI-driven arbitrator research, profiling and conflict checks.

Considerations when using AI in legal and arbitrator research

While AI tools may facilitate legal and arbitrator research, they require careful use. A key consideration is the potential for inaccuracies and hallucinations. Outputs must therefore be checked to avoid inaccurate submissions. 

A related consideration is the risk of over-reliance on AI. AI outputs appear polished and are produced in seconds. This may create an impression of reliability and completeness that tempts users to delegate the entirety of their case or arbitrator research. Legal analysis, however, depends on logical reasoning and sound judgment. It is crucial to know and understand why and how a reference is being used ̶ whether to support a legal argument or to determine whether a candidate should be considered as a potential arbitrator. AI systems, however, do not disclose how their outputs are obtained. Even if prompted to explain, the responses themselves may contain errors or hallucinations. A further consideration is that, over time, over-reliance may lead to a decline in critical thinking skills. 

AI use also calls for careful attention to avoid inadvertent disclosure of confidential client information or breach of data privacy regulations. Users of AI must therefore ensure that AI tools provide confidentiality safeguards, store data in jurisdictions with adequate privacy protections, process personal information only to the extent necessary to pursue legitimate interests such as ensuring that a potential arbitrator has the necessary qualifications and assessing their independence and impartiality, do not improperly train on user inputs, and protect data from security breaches.

AI use and a lawyer’s professional responsibilities

Recognising the increasing use of AI by the legal industry, some bar associations have issued guidance on whether and how professional standards and obligations apply. For example:

• The American Bar Association’s (ABA) Formal Opinion 512 advises that the duty to provide competent representation means that lawyers have a continuing obligation to maintain a “reasonable understanding” of GenAI’s limits and capabilities. Lawyers are also under a duty not to knowingly make a false statement of law or fact. Reliance on AI without appreciating its propensity for error may breach these duties. Meanwhile, the duty of maintaining the confidentiality of client information means lawyers must assess the risks of disclosure and unauthorised access whether to the public or to other persons within the same firm before uploading client data into any AI tool. 
• The French Conseil National des Barreaux’s (CNB) March 2026 guide on Professional Ethics and Artificial Intelligence warns that the duty to maintain confidentiality of client information and comply with data privacy regulations may require ensuring AI tools are not trained on client data and that client and third party personal information are anonymised in uploads and prompts. On the other hand, the duty of competence requires that lawyers master the AI tools they use, to maintain that expertise through regular training and to avoid using AI to provide advice on legal matters for which the lawyer otherwise lacks expertise. The duty of diligence and prudence means that lawyers must verify the AI’s results before passing it on to the client, court or tribunal. Relatedly, the CNB opines that the duty of maintaining intellectual independence requires lawyers to avoid blind reliance on AI, whose output may reflect biases in training data. 
• England and Wales’ Solicitors Regulation Authority and the Law Society have confirmed that the use of AI is subject to the SRA’s standards, principles and Code of Conduct. More specifically, the High Court in Re an Office-Holder; Cork v Smith and Ayinde v London Borough of Haringeyhas stressed that the use of AI in legal research gives rise to a professional duty to check its accuracy. Importantly, Ayinde warns that blind reliance on AI-generated citations can breach core duties: advancing only arguable submissions, citing relevant authorities, providing competent advice, and not misleading the court. Sanctions may vary from public admonition, the imposition of a costs order, contempt proceedings and even criminal investigation. 

Users of AI in the EU, particularly arbitrators, may also do well to familiarize themselves with the EU AI Act. While the Act does not prohibit the use of AI in dispute resolution, it classifies certain uses of AI in alternative dispute resolution as high-risk under Annex III(8)(a). This includes AI use by adjudicators, including arbitrators, to carry out legal research and analysis. This classification gives rise to a range of obligations, including ensuring human oversight, risk management, record-keeping, technical documentation and data governance. These obligations will apply from 2 December 2027.

AI and legal research

With those caveats in mind, the benefits of AI for legal research are clear: AI does not simply accelerate the old research process, it allows practitioners to move more quickly from collection and analysis to building legal strategy.

In particular, AI can help identify and define legal issues, map relevant authorities, generate case summaries, compare awards and judgments, and surface patterns that may otherwise remain buried in the material. It can trace the history of a principle across a line of cases, identify where tribunals have converged or diverged, expose gaps in the authorities, and assemble the results into a first-cut research memorandum in a matter of minutes. Properly directed, it can then stress-test that memorandum: checking whether the analysis has missed a material authority, an argument has been underdeveloped or if a quotation is properly supported.

Beyond case retrieval and analysis, agentic AI may also be used as a “partner” in improving the research process itself. Legal practitioners can use AI to brainstorm, generate and refine research scope and strategy, identify and pursue relevant searches, suggest and research applicable jurisdictions and propose analytical frameworks tailored to the specific legal question – subject always to the professional and ethical duties outlined above.   

AI and arbitrator research

AI may also assist in arbitrator research and selection. It can help define the selection criteria for an ideal arbitrator, extract and aggregate information on arbitrator candidates and their areas of expertise, whether there are any trends on their appointments by respondent, claimants or as chair, identify possible conflicts of interest and perform candidate comparisons. 

Existing AI tools are already capable of profiling judges, analysing their positions on specific legal issues and patterns of reasoning by reviewing prior judgments and the extent to which their decisions have been upheld on appeal. AI has even been used to predict a judge’s openness or tendency to adopt certain arguments, the likely issues and questions they may raise, the arguments they may focus on and how they may decide a case or respond to an interim application. 

While arbitration-specific platforms do not yet offer this level of analysis, depending on the arbitrator’s profile and the breadth of information available on their cases and awards, it may in due course be possible to carry out a similar analysis using a combination of arbitration-specific and general-purpose agentic AI. Profiles of former judges who have moved into arbitration may be especially suitable for this kind of analysis: their judgments can provide a substantial public record from which AI can help identify patterns of reasoning and decision-making tendencies (while, of course, appreciating that there will always be limits to the utility of this sort of analysis).

In sum, the use of AI in legal and arbitrator research is likely to become increasingly routine. As the technology continues to develop, the key question is no longer whether AI will be used in legal and arbitrator research, but how it can be deployed responsibly to improve the quality, speed and strategic focus of legal analysis while remaining consistent with professional and ethical obligations.