Regulation (EU) 2024/1689 (the “AI Act”), approved by the Council of the European Union on 21 May 2024[1] and entered into force on 1 August 2024,[2] is a regulation on artificial intelligence (“AI”) that establishes extensive legal requirements in relation to AI systems classified as “high-risk”.[3]
In support of the AI Act, the European Commission published a set of draft guidelines on the classification of high-risk AI systems (the “Draft Guidelines”) on 19 May 2026, which aim to set out the European Commission’s interpretation of certain concepts relevant for the classification of AI systems under the AI Act, along with practical examples of AI systems that should or should not be classified as high-risk, thereby facilitating the uniform application and effective enforcement of the Act.[4]
Stakeholders, including providers and developers of AI systems, businesses and public authorities as well as academia, research institutions and citizens, were initially invited to share their views by 23 June 2026; the consultation period was subsequently extended to 23 July 2026.[5]
For arbitration practitioners, the Draft Guidelines raise a particularly sensitive question: when AI is used in arbitral proceedings, who is the regulated actor – the arbitral institution that administers the case, the tribunal that decides it, or both?
Therefore, in a letter dated 23 July 2026 (the “Joint Letter”), sixteen of the world’s leading arbitral institutions, including the International Chamber of Commerce (“ICC”) International Court of Arbitration & International Centre for ADR, the International Centre for Dispute Resolution (“ICDR”) / American Arbitration Association (“AAA”), the International Centre for Settlement of Investment Disputes (“ICSID”), the London Court of International Arbitration (“LCIA”), the Chartered Institute of Arbitrators (“CIArb”) and the SCC Arbitration Institute, responded to the European Commission’s invitation, warning that the Draft Guidelines risk imposing regulatory obligations on the wrong actors and obscuring where accountability for AI-assisted decision-making should properly lie.[6]
This note examines how the AI Act and the Draft Guidelines may apply to the use of AI in arbitration, the concerns raised in the institutions’ joint response, and the potentially significant consequences for arbitral institutions, tribunals and technology providers.
The AI Act
At the heart of the AI Act, which applies not only to AI actors established in the EU, but also to certain non-EU providers and deployers where their AI systems or outputs are used in the EU,[7] is a simple but significant idea: the more serious the potential impact of an AI system, the more demanding the regulatory framework that applies to it. The Act therefore adopts a risk-based approach, distinguishing between prohibited practices, high-risk systems and lower-risk applications.
For high-risk AI systems, the AI Act lays out specific requirements that must be fulfilled, such as the implementation of a risk management system[8] and data governance and management practices,[9] maintaining up-to-date technical documentation[10] and record-keeping,[11] and employing human oversight.[12]
Article 6 of the AI Act provides that AI systems should be considered high-risk when, inter alia, they are referred to in Annex III to the AI Act. However, an AI system referred to in Annex III shall not be considered to be high-risk where it “does not pose a significant risk of harm to the health, safety or fundamental rights of natural persons, including by not materially influencing the outcome of decision making.”[13]
Annex III, in turn, presents a list of specific areas, such as biometrics, education and vocational training, and law enforcement, in which AI systems will be considered high-risk.
Relevantly, point 8 of Annex III labels as “high-risk” AI systems employed for the purposes of the “[a]dministration of justice and democratic processes”, including “AI 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”.[14]
This provision is therefore the principal gateway through which AI used in arbitration may fall within the AI Act’s high-risk framework.
The Draft Guidelines
The European Commission’s Draft Guidelines, which are intended to clarify the classification of AI systems as high-risk pursuant to the AI Act, attempt, inter alia, to shed light on the kinds of uses in alternative dispute resolution (“ADR”) that fall under point 8 of Annex III.
First, the Draft Guidelines explain that examples of ADR include “arbitration, mediation, conciliation, ombudsmen and complaints boards”.[15]
They then provide that for an AI system intended to be used in ADR to be classified as high-risk, two conditions must be fulfilled:
- The system must be intended to be used by an ADR body where the outcome of the ADR proceedings produces legal effects for the parties; and
- The intended purpose of the system must be to assist the ADR body in researching and interpreting facts and the law and in applying the law to a concrete set of facts.[16]
The Draft Guidelines define “ADR bodies” as including “a wide set of entities that deal with civil, including commercial and labour, and other disputes”, such as “[c]ommercial arbitration institutions (e.g. national and international arbitration bodies) or investment dispute bodies that resolve disputes between businesses”.[17]
With respect to “producing legal effects for the parties”, the Draft Guidelines specify that arbitral awards may typically be so classified as they generally have binding and final effect on the parties, are enforceable under national law and are internationally recognised through the United Nations Convention on the Recognition and Enforcement of Foreign Arbitral Awards (the “New York Convention”).[18]
The Draft Guidelines then set out several practical examples of AI systems that would be classified as high-risk in this context. These include AI systems that:
- Analyse the factual circumstances of the case, submissions received, identify applicable law and case law and then generate drafts of judicial decisions/judgments or their parts, in particular the reasoning part and decision;[19]
- Generate (draft) decisions in small claims cases or orders for payment, relying on structured case data and legal templates;[20]
- Are intended to support judges in handling identical or similar cases by extracting relevant facts from the claim and clustering them based on recurring factual constellations across similar cases, researching a database of previous decisions to find similar cases and identify legal reasoning patterns, and preparing suggestions and corresponding text modules for the final decision based on outcomes of similar previous cases;[21] and
- Analyse the presented facts of the case (or their summary), then assist judges in identifying relevant laws and legal precedents and suggest how they apply to the facts of a case.[22]
The Draft Guidelines also provide examples of certain uses of AI systems that fall outside the high-risk classification under Annex III, point 8, including speech-to-text systems, AI systems facilitating communication with the public, case assignment systems, processing and management of evidentiary submissions in judicial proceedings, and other ancillary administrative activities.[23]
They further note that uses such as assisting in pre-classification of incoming applications or claims, metadata extraction, advanced search engines, language editing assistance and suggesting factual questions not addressed in a draft are exempted by the filter mechanism in Article 6(3) of the AI Act.[24]
The Joint Letter
The arbitral institutions put forward several critiques of the Draft Guidelines in their Joint Letter.
First, they pointed out that the Draft Guidelines misattribute arbitral decision-making: “Because arbitral institutions administer proceedings and do not decide the merits of the dispute […], treating arbitral institutions as the relevant ‘ADR body’ for the purposes of Annex III, point 8(a) of the AI Act risks attaching the resulting obligations to a wrong actor.”[25][26]
In this same vein, the institutions requested that, as the Draft Guidelines place judicial administration (defined as “institutions or bodies that are responsible for the management, governance, and support of the judiciary, rather than for the adjudication of cases themselves”) outside the scope of the AI Act,[26] a similar exclusion should apply to arbitral institutions, whose role is analogous.[27]
Finally, the arbitral institutions suggested that the European Commission consider the implications of the inclusion of the terminology “investment dispute bodies that resolve disputes between businesses”, as investment disputes are typically between investors, which may be businesses or individuals or both, as claimants and States as respondents.[28]
According to a statement by the SCC, “[t]he joint submission represents a significant and unified intervention from the international arbitration community.”[29]
The Joint Letter therefore seeks to align responsibility for AI-assisted decision-making with the arbitral tribunal, as the body exercising the adjudicative function, rather than with the institution administering the proceedings.
Implications
It is clear that the AI Act could have a direct impact on the use of AI in arbitration. What remains considerably less clear is how the relevant obligations would apply in practice, and in particular which actors within the arbitral process would bear them.
As pointed out in the Joint Letter, arbitral institutions appear to fall squarely within the Draft Guidelines’ broad definition of “ADR bodies”. Yet those institutions do not ordinarily decide the merits of disputes or render arbitral awards producing legal effects for the parties; that function is performed by the arbitral tribunal.
This creates a potential disconnect between the entity identified by the Draft Guidelines as the relevant ADR body and the actor actually responsible for substantive decision-making. It is also unclear whether an arbitrator using AI independently would fall within the relevant framework, particularly in ad hoc arbitration where no arbitral institution is involved.
This concern lies at the heart of the Joint Letter. As the SCC noted in a recent statement, the Draft Guidelines as currently formulated would “create disproportionate regulatory burdens and risk obscuring where genuine accountability for AI-assisted decision-making in arbitration should rest: with the arbitral tribunal.”[30]
If the European Commission were to accept the institutions’ proposed distinction between arbitral administration and adjudication, arbitral institutions acting in a purely administrative capacity would likely fall largely outside the scope of the relevant high-risk provisions. Responsibility would instead more closely track the use of AI in the substantive determination of disputes.
If, however, the Draft Guidelines remain unchanged, arbitral institutions may face greater compliance uncertainty and potentially significant regulatory burdens. Institutions administering EU-connected arbitrations could be required to consider whether AI tools used within proceedings fall within the high-risk framework and, depending on how the Act is interpreted, whether safeguards are required in relation to AI use by arbitrators acting under their rules. That could increase administrative complexity and compliance costs and, ultimately, the cost of arbitration for the parties.
Even if the Commission accepts the institutions’ broader argument that purely administrative functions should fall outside the Act, a more difficult boundary question remains. Arbitral institutions do not perform only administrative tasks. They may also take procedural decisions that can materially affect the conduct of proceedings. It is therefore unclear whether the use of AI by an institution to research or draft such decisions would fall within the scope of the Act under the Draft Guidelines.
For example, an arbitral institution like the ICC Court might hypothetically use AI to search prior Court decisions, identify recurring factual or reasoning patterns, and apply those patterns to the procedural circumstances of a particular case. Decisions such as whether the parties have prima facie agreed to arbitrate or whether a challenge to an arbitrator should be upheld may materially affect the parties’ procedural position and may, in turn, indirectly influence the ultimate outcome of the arbitration. The difficult question is whether AI used in that context would remain merely administrative, or whether it would be used in a manner sufficiently analogous to the research, interpretation and application of law contemplated by Annex III and the Draft Guidelines.
Further, some emerging uses of AI in arbitration raise a different set of questions because they involve AI more directly in the substantive decision-making process. One example is the AAA’s AI Arbitrator. Where the parties agree to its use, the system undertakes claim parsing, document summarisation and analysis, applies legal reasoning, and generates a draft recommended award, which is then reviewed, finalised and issued by a human arbitrator.[31]
Whether an arbitration system of this kind falls within the scope of the AI Act is not clear. Its functions bear resemblance to the types of AI use identified in the Draft Guidelines, particularly systems that analyse the facts of an individual dispute, identify relevant legal reasoning, and assist in preparing a draft decision. At the same time, the role of the human arbitrator in reviewing and finalising the award, as well as the precise legal characterisation of the AAA and the AI system’s role within the proceedings, may be relevant to the classification analysis.
The AAA example therefore illustrates a broader unresolved question: where AI moves beyond administrative support and begins to contribute to the reasoning or drafting underlying an arbitral award, at what point does its use become sufficiently connected to the decision-making process to fall within the high-risk framework? The Draft Guidelines provide some indications, but do not appear to resolve that question definitively.
Conclusion
The Draft Guidelines provide clarification on how the AI Act may apply to arbitration, but they also expose certain difficulties in applying a regulatory framework designed around judicial decision-making to the arbitral context. Arbitral institutions, tribunals and individual arbitrators perform distinct functions, and failing to recognise these distinct functions risks obscuring where responsibility for AI-assisted decision-making should properly lie.
The Joint Letter thus raises valid concerns. A clearer distinction between the administrative role of arbitral institutions and the adjudicative role of tribunals would better reflect how arbitration operates in practice. At the same time, that distinction may not resolve every case. Institutional procedural decisions, as well as tools that involve AI more directly in analysing claims or assisting with the preparation of awards, raise harder questions about where administration ends and decision-making begins.
Much will therefore depend on how the European Commission addresses these issues in the final Guidelines. Greater clarity will be useful not only for arbitral institutions and tribunals seeking to understand their compliance obligations, but also for technology providers developing tools for use in arbitration. As AI becomes increasingly integrated into arbitral practice, the challenge will be to ensure that the AI Act captures uses that genuinely influence legal decision-making without unnecessarily extending high-risk obligations to administrative functions that do not.
[1] Future of Life Institute, Historic Timeline, https://artificialintelligenceact.eu/developments/ (last accessed 13 August 2026).
[2] Future of Life Institute, Implementation Timeline, https://artificialintelligenceact.eu/implementation-timeline/ (last accessed 13 August 2026).
[3] Future of Life Institute, The EU Artificial Intelligence Act: Up-to-date developments and analyses of the EU AI Act, https://artificialintelligenceact.eu/ (last accessed 13 August 2026).
[4] European Commission Policy and Legislation, Draft Commission guidelines on the classification of high-risk AI systems, 19 May 2026, https://digital-strategy.ec.europa.eu/en/library/draft-commission-guidelines-classification-high-risk-ai-systems (last accessed 13 August 2026).
[5] Ibid; European Commission, Targeted consultation on the draft guidelines for the classification of high-risk artificial intelligence systems, 19 May 2026, https://digital-strategy.ec.europa.eu/en/consultations/targeted-consultation-draft-guidelines-classification-high-risk-artificial-intelligence-systems (last accessed 13 August 2026).
[6] SCC Arbitration Institute, SCC Arbitration Institute Joins Global Coalition in EU AI Act Consultation, 24 July 2026, https://sccarbitrationinstitute.se/en/news/scc-arbitration-institute-joins-global-coalition-in-eu-ai-act-consultation/ (last accessed 13 August 2026).
[7] Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence and amending Regulations (EC) No 300/2008, (EU) No 167/2013, (EU) No 168/2013, (EU) 2018/858, (EU) 2018/1139 and (EU) 2019/2144 and Directives 2014/90/EU, (EU) 2016/797 and (EU) 2020/1828 (Artificial Intelligence Act) [2024] L2024/1689 (“AI Act”), Art. 2.
[8] Id., Art. 9.
[9] Id., Art. 10.
[10] Id., Art. 11.
[11] Id., Art. 12.
[12] Id., Art. 14.
[13] Id., Art. 6(3).
[14] Id., Annex III, pt. 8.
[15] Draft Commission guidelines on the classification of high-risk AI systems under Article 6 of Regulation (EU) 2024/1689 (AI Act) for stakeholder consultation [2026] (“Draft Guidelines”), para. 426.
[16] Id., para. 425.
[17] Id., para. 429.
[18] Id., para. 431.
[19] Id., p. 135.
[20] Id., p. 136.
[21] Ibid.
[22] Ibid.
[23] Id., pp. 136-137.
[24] Id., p. 137.
[25] Joint Letter Re: Targeted consultation on the draft Commission Guidelines on the classification of high-risk AI systems (Article 6 of Regulation (EU) 2024/1689) – Joint submission of international arbitration institutions dated 23 July 2026 (“Joint Letter”), p. 2.
[26] Draft Guidelines, para. 411.
[27] Joint Letter, p. 2.
[28] Ibid.
[29] SCC Arbitration Institute, SCC Arbitration Institute Joins Global Coalition in EU AI Act Consultation, 24 July 2026, https://sccarbitrationinstitute.se/en/news/scc-arbitration-institute-joins-global-coalition-in-eu-ai-act-consultation/ (last accessed 13 August 2026).
[30] Ibid.
[31] American Arbitration Association, AI Arbitrator: A New Path to Dispute Resolution, https://www.adr.org/media/1vva0v2y/ai-arbitrator-one-sheet-20251029-1.pdf (last accessed 13 August 2026); American Arbitration Association, AI Arbitrator, https://www.adr.org/ai-arbitrator/ (last accessed 13 August 2026).
