The EU AI Act's draft guidelines on high-risk AI systems have triggered a coordinated protest from sixteen of the world's leading arbitral institutions. On 23 July 2026, bodies including the ICC International Court of Arbitration, ICDR/AAA, ICSID, the LCIA, the CIArb and the SCC Arbitration Institute jointly told the European Commission that the rules risk placing regulatory obligations on the wrong actors in arbitration proceedings.
The dispute centres on who should be accountable when AI assists in dispute resolution. The draft guidelines, published on 19 May 2026, classify AI systems used in arbitration as high-risk where the system's intended purpose is to assist in resolving disputes and the outcome has legal effects on the parties. Because arbitral awards are generally binding, final and enforceable under the New York Convention, AI tools that assist in drafting decisions are likely to fall within this category.
The AI Act, in force since 1 August 2024, imposes obligations on providers and deployers of high-risk systems, including risk management, data governance, technical documentation, record-keeping and human oversight. Under Article 6 and Annex III, AI used "by a judicial authority or on their behalf to assist in researching and interpreting facts and law and applying the law to a concrete set of facts, or used in a similar way in alternative dispute resolution" is considered high-risk.
The institutions' objection
The sixteen institutions argue that the guidelines misattribute responsibility. Their core contention: arbitral institutions do not decide cases; tribunals do. Imposing obligations on institutions under the high-risk framework misunderstands the adjudicative process, they said, and risks burdening actors who exercise no decision-making authority over outcomes.
The institutions urged the Commission to distinguish between administrative support and adjudicative functions. They asked for clarification that institutional case administration should fall outside the AI Act's high-risk classification, just as judicial administration does.
The distinction is not always clean. Institutions sometimes issue procedural decisions that affect the outcome of arbitration, such as decisions on arbitrator challenges or prima facie jurisdictional determinations. If AI assisted in those functions, it could cross the line from administration into decision-making support. The AAA's AI-assisted arbitration tool, which generates draft awards after analysing facts and case law, raises the same question in practical form.
Where responsibility should rest
The joint letter reflects the arbitration community's position that responsibility for AI-assisted decision-making should sit with those exercising adjudicative functions - tribunals and arbitrators - rather than with the institutions that administer proceedings. The Commission invited stakeholder comments by 23 July 2026, and the final guidelines are expected to clarify the issue.
For legal professionals working with AI in dispute resolution, the outcome will determine compliance obligations. If the Commission accepts the institutions' position, arbitrators and tribunals will bear the regulatory burden directly. If it does not, institutions may need to implement high-risk compliance frameworks for tools they use in case administration, even where they have no role in deciding the merits.
Either way, the practical question for practitioners is the same: which AI tools in your workflow touch decision-making, and can you demonstrate human oversight over them? That question applies whether you sit on a tribunal, administer cases, or advise parties. The AI for Legal resources track these developments as they take shape. For those supporting arbitration teams, the AI Learning Path for Paralegals covers the compliance fundamentals now becoming central to institutional practice.
Why this matters for legal professionals
The final guidelines will determine who must implement risk management, data governance and human oversight measures when AI assists in arbitration. The institutions' joint response signals that the arbitration community expects accountability to follow adjudicative authority, not administrative function. Legal professionals should map their own AI usage against that principle now, before the compliance requirements are locked in.
Your membership also unlocks:
