The EU AI Act and International Arbitration

The EU AI Act is now officially in place, and after this year’s updates, it brings transformative implications that every international arbitrator must understand. The Act expressly classifies certain AI use in dispute resolution as "high-risk", carrying obligations, transparency duties, and fines of up to €15 million for non-compliance - which means that it’s more important than ever for all legal professionals to understand how it affects them. 

That’s why we’ve created this article - to walk you through the EU AI Act and its requirements, and how purpose-built tools like TrialView can help you work confidently within its guidelines. 

Background

The EU AI Act (Regulation (EU) 2024/1689) was implemented on the 1st August 2024[1]. It is the world's first comprehensive framework for AI regulation, built around a tiered risk classification system. The most significant section for arbitrators is Annex III, which lists categories of "high-risk" AI use cases.

In May 2026, the European Parliament and Council reached a provisional agreement on the EU Digital Omnibus on AI - a package of targeted amendments to the Act. Under that agreement, the high-risk AI obligations for Annex III systems have been deferred from the 2nd August 2026 to the 2nd December 2027. 

However, the Article 50 transparency obligations, which require organisations to inform individuals when they are using an AI system, remain on their original schedule and apply from the 2nd August 2026.[2] Arbitrators should treat this as an immediate focus, while the broader Annex III high-risk framework should initiate some active preparation with a plan to be ready by the extended deadline.[3]

Something to bear in mind is that the Act is not limited to the EU. It applies to any AI systems established or located in the EU, but also outside of the EU when the output of the AI system is used within the Union. In practice, this captures a very wide range of international arbitration. Individual arbitrators are also AI ‘deployers’ within the meaning of Article 3(4), which defines a deployer as any natural or legal person using an AI system under their authority. So are arbitral institutions, in respect of the AI tools they use in administering proceedings.

How is it relevant to international arbitration?

Annex III, point 8(a) of the Act classifies any AI system "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" as high risk.[4] The Draft Guidelines (p. 139) confirm that this is applicable to arbitration proceedings as well, not just judicial authorities. [5]

Recital 61 clarifies further that AI systems used by alternative dispute resolution bodies are high-risk "when the outcomes of the alternative dispute resolution proceedings produce legal effects for the parties." and that “The use of AI tools can support the decision-making power of judges or judicial independence, but should not replace it: the final decision-making must remain a human-driven activity.” Therefore, any legal professional that is currently using AI to assist in their cases is affected by the Act and specifically this recital. 

What counts as high-risk?

As shown in Recital 61, the Act does not actually prohibit AI use in arbitration. It creates a classification system containing different requirements depending on the nature of the AI use. The key distinction is between AI that assists in the arbitration procedures and AI that performs purely preparatory, procedural, or administrative functions. Recital 61 states “The classification of AI systems as high-risk should not, however, extend to AI systems intended for purely ancillary administrative activities that do not affect the actual administration of justice in individual cases, such as anonymisation or pseudonymisation of judicial decisions, documents or data, communication between personnel, administrative tasks.”

AI that assists an arbitrator in researching and interpreting facts and law, or in applying the law to a concrete set of facts, automatically falls within the high-risk category. But different tools will also sit at different points along this spectrum.

Article 6(3), provides a significant category of AI use that is not considered high-risk, even where it assists in an Annex III context, because it does not materially influence the outcome. 

What can you do without triggering the high-risk classification?

Calculating deadlines, processing advance payments, and scheduling fall within the narrow procedural task exemption under Article 6(3)(a), purely administrative functions carrying no decision-making weight. Formatting a finalised procedural order, reformatting a prepared chronology for clarity, or style-checking a draft award are examples of "improving the result of a previously completed human activity" under Article 6(3)(b).

Generating a chronology of facts with source links, creating a summary of the evidence record, or searching for case law and literature on a legal issue are preparatory tasks to a human assessment under Article 6(3)(d). They prepare the ground work but the arbitrator still creates the legal strategy. 

Similarly, flagging where an award may deviate from the tribunal's prior procedural decisions falls within the pattern-detection exemption under Article 6(3)(c), so long as the AI is not meant to replace or influence the previously completed human assessment. Transcription, translation, document anonymisation, and advanced search across case materials are explicitly listed in the Draft Guidelines as falling outside of the high-risk classification.

Using high-risk AI

Even if a task or tool is classified as high-risk AI, you are still able to use them as long as you meet a certain set of obligations. Article 26 of the Act sets out the obligations of deployers and for arbitrators, it is about providing human oversight, log retention, transparency to parties, and provider due diligence.

Human oversight for example requires appropriate technical and logistical measures to be in place, with a competent, trained individual monitoring the system's use. Logs should be automatically generated by the high-risk system and must also be retained for at least six months. 

Another consideration is that under Article 50, which applies from the 2nd August 2026, people must be informed when they are interacting with an AI system. This sits alongside the Article 26(4) obligation to inform affected individuals that a high-risk AI system has been used, a disclosure obligation also recommended by the Silicon Valley Arbitration and Mediation Center (SVAMC) Guidelines on AI in Arbitration.[7] Arbitrators should also verify that providers of high-risk AI systems hold the required technical documentation, declaration of conformity, and EU database registration.

Penalties for breach are reserved for Member States to specify, but the Act sets a ceiling of €15 million for violations of Article 26 deployer obligations, with the same applied to Article 50 transparency violations.

How TrialView can help

TrialView is an AI-powered platform for disputes trusted by leading law firms, courts, and arbitral institutions (including the ICC, IAC, and DDRC). Our AI finds and surfaces information from the case portfolio but leaves it to the user to create their legal strategy, aligning us naturally with the compliance framework in the EU AI Act. 

Our AI features are built as preparatory and presentational tools - they organise, retrieve, surface, and flag, but they do not decide. We believe that our tool makes a good lawyer great, by providing them with the information they need to develop a sharp legal strategy and apply their expertise in the right way in the courtroom.

For example, TrialView’s automated bundle creation generates court-compliant bundles with automatic pagination, date recognition, indexing, and cross-referencing. And our AI-powered document analysis allows arbitrators and counsel to query the entire case record (not just a few pages) in natural language, with the system returning sourced and hyperlinked references for review.

The platform’s timeline generation feature also automatically builds detailed, hyperlinked chronologies from case materials. And our latest Case Intelligence capabilities with real-time transcription identify patterns and potential inconsistencies in witness statements and evidence, flagging them for human review. 

All of this takes place within a secure, encrypted workspace where case data is never used to train external models, supporting both the confidentiality obligations fundamental to arbitration and the data governance expectations of the Act.

Practical guidance

The Act is not a reason to avoid AI. But it is a reason to ask more questions about the AI you use and how it is implemented. Before deploying any AI tool in an arbitral context, four questions are worth asking: 

• What would you like the tool to do? What are your pain points?
• Does the tool assist you in finding and organising information or does it form views? 
• If the tool falls within the high-risk category, have you verified the provider's credentials, arranged log retention, and disclosed use to the parties? 
• Are you applying your own judgment, informed by the tool, or justifying a decision the tool has already made?

The EU AI Act brings clarity and security to AI as we navigate its use in our industry going forward. The framework it establishes is workable, and the recent AI Omnibus provisional agreement provides additional time to prepare for the high-risk obligations.

But what the Act firmly prevents is the use of AI as a decision-maker. The personal mandate of the arbitrator is non-delegable. AI can inform, surface, and prepare, but the judgment must remain human.

TrialView sits on the right side of these requirements, and for practitioners navigating this new regulatory landscape, purpose-built legal AI tools that operate transparently and securely are both legally and operationally superior to general-purpose alternatives.

References

1. EU AI Act (Regulation (EU) 2024/1689), Official Journal of the European Union, 12 July 2024. https://artificialintelligenceact.eu
   
   
2. Sidley Austin, "EU AI Act Transparency Obligations: Preparing for Compliance by 2 August 2026", Data Matters, 24 June 2026. https://datamatters.sidley.com/2026/06/24/eu-ai-act-transparency-obligations-preparing-for-compliance-by-2-august-2026
   
   
3. European Commission, Digital Strategy: AI Act. https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai
   
   
4. EU AI Act, Annex III, point 8(a). https://artificialintelligenceact.eu/annex/3
   
   
5. EU AI Act Draft Commission Guidelines on the Classification of High-Risk AI Systems under Article 6 ("Draft Guidelines"), p. 139. https://www.twobirds.com/en/insights/2026/the-commission's-draft-high-risk-ai-guidelines-under-the-eu-ai-act-a-first-read
   
   
6. Draft Guidelines, Annex III, p. 135 and paragraph 419. https://www.modulos.ai/blog/eu-ai-act-annex-iii-draft-guidelines-what-changed
   
   
7. Heetkamp, Beimel and Lüttenberg, "How Does the EU AI Act Apply to Arbitration?", SchiedsVZ / German Arbitration Journal, 2024, pp. 225-228. https://dailyjus.com/legal-tech/2025/01/how-does-the-eu-ai-act-apply-to-arbitration
   
   
8. Kluwer Arbitration Blog, "We Need to Talk About the EU AI Act", 2025. https://legalblogs.wolterskluwer.com/arbitration-blog/we-need-to-talk-about-the-eu-ai-act
   
   
9. SVAMC, "Guidelines on the Use of Artificial Intelligence in Arbitration", April 2024. https://svamc.org/svamc-publishes-guidelines-on-the-use-of-artificial-intelligence-in-arbitration/