AI in dispute resolution in Europe: the rules, courts and institutions

The short version

• European courts use AI mostly for support work. A 2025 report by the AI Advisory Board of the Council of Europe’s European Commission for the Efficiency of Justice (CEPEJ) found no fully autonomous AI system in courts.
• Every court text we checked keeps the decision on the case with a judge.
• From 2 December 2027, the EU AI Act can treat AI that helps a court or an alternative dispute resolution (ADR) body decide a case with legal effects as high-risk. The European Commission’s draft guidelines of 19 May 2026, not yet adopted, put court website chatbots, transcription, case assignment and anonymisation outside that category.
• Under national rules applying from 20 September 2028, EU consumer ADR bodies that use automated means in decision-making must tell parties in advance and let them ask for review by a natural person.
• The EU’s General Data Protection Regulation (GDPR) already restricts decisions based solely on automated processing. The Council of Europe’s AI treaty is not yet in force.
• On 23 July 2026, 16 arbitral institutions and bodies asked the Commission to keep their administrative tools outside the high-risk label.^{2,4,5,6,8,15,20,22,23,24,37,53}

How is AI used in dispute resolution in Europe?

AI in dispute resolution in Europe is mostly support work around a human decision: anonymising, transcribing, sorting mass claims and drafting under a judge’s control. This guide covers European Union (EU) law, Council of Europe texts and national court rules on artificial intelligence in courts, arbitration and other alternative dispute resolution (ADR). UK courts and bodies are included, although the UK is not in the EU.^14,15

The CEPEJ’s Resource Centre on Cyberjustice and AI focuses on public-sector systems. By the end of 2024 it had listed 125 tools “primarily from Europe”, 41 of them for decision support and 7 for online dispute resolution, according to the first report of its AI Advisory Board (AIAB), February 2025 (paras 8, 25, 41). The board found no fully autonomous AI system in courts (para 10). In an October 2025 CEPEJ survey, 46% of the 33 responding member states reported generative AI use in courts, mostly for summarising (CEPEJ(2025)18Final, para 10).^14,15,16

A Stuttgart appeal court’s assistant classifies thousands of diesel-emission claims (AIAB report, p. 8). A Frankfurt local court initiated a tool, run by the Hesse justice ministry, for its roughly 15,000 air-passenger claims a year. In the 2023 prototype, the judge picks the outcome, then the tool proposes editable facts and reasons (Hesse justice ministry, 13 November 2023). Spain’s justice ministry built a tool for the Supreme Court that “does not decide, resolve or propose judgments” (1 October 2026). In 2024 Sweden’s courts began using AI to suggest what to mask before documents are released, and a judge reviews each document (21 October 2024).^15,26,33,34

Outside the courts, in the 2025 audit by the Centre for Effective Dispute Resolution (CEDR), 59% of responding UK mediators had used AI. The UK’s Financial Ombudsman Service uses AI to support its service while people decide cases. Our dispute resolution guide covers mediation, online dispute resolution and ombudsman schemes. Our other regional guides cover Australia and South East Asia.^56,57

When does the EU AI Act treat AI in courts and ADR as high-risk?

From 2 December 2027, the EU AI Act can treat AI that helps a court or an ADR body decide a case with legal effects as high-risk. Regulation (EU) 2026/1744, the Digital Omnibus on AI, entered into force on 27 July 2026 and set that date (art. 1 point (40)(b)).^1,2,3,4

Annex III of the Act lists high-risk uses. Its point 8(a) covers AI intended to be used by or for a judicial authority to assist it 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”. Recital 61, in the Act’s preamble, limits that ADR part to outcomes with legal effects and leaves out purely ancillary administrative activities.²

The European Commission’s draft guidelines of 19 May 2026, not yet adopted, explain the court use case. Point 8(a) holds “two distinct use cases”, one for courts and one for ADR (para 405). It does not reach judicial administration, meaning the bodies that manage and support the courts rather than decide cases (para 411), or tools used by parties and their lawyers (para 416). Helping either to research and interpret facts and law or to apply the law to the facts is enough (paras 417, 418). The draft sorts court tools into three groups (printed pp. 135 to 139):^2,5

• Outside point 8(a): speech-to-text, court website chatbots, press summaries, case assignment, chronologies, anonymisation and plain keyword or database search (para 421).
• In scope but exempt under art. 6(3), which covers tools that do not materially influence the outcome: pre-classifying claims, as a narrow procedural task or preparatory task; a public case-law database that also summarises decisions and searches in natural language; proofreading.
• In scope, with the art. 6(3) exception ruled out: tools that draft decisions in small claims and payment-order cases, and tools that find the relevant law and past rulings for the facts of a case and propose how they fit.^2,5

The draft lists assigning cases, dispatching documents, verifying court fees and anonymising decisions as ancillary administrative tasks (printed pp. 137, 138). On our reading, case management software that does only these sits outside point 8(a). A tool that recommends whether a claim is admissible, beyond technical checks, cannot rely on the art. 6(3) exceptions (printed p. 138).⁵

The court examples also guide ADR (para 434). Arbitral awards generally bind the parties (para 431). AI used where the outcome is a voluntary agreement or non-binding recommendation, as in most mediation, is outside point 8(a) (paras 430, 431). Our dispute resolution guide covers how the AI Act and the consumer ADR directive apply to mediation and ombudsman schemes. Our arbitration guide covers how the AI Act applies to arbitration, including the art. 6(3) exception, deployers’ duties and its reach to seats outside the EU.⁵

Courts that use a high-risk system are public-body deployers. From 2 December 2027 they must assess its impact on fundamental rights before use (art. 27(1)). That duty falls on bodies governed by public law and private entities providing public services, not on every deployer. The assessment may cross-refer to their data protection impact assessment (art. 27(4), as replaced by art. 1 point (13)(a) of the Omnibus). Courts must also tell people when such a system is used in decisions about them (art. 26(11)).^2,4

Public-authority high-risk systems already on the market or in service before 2 December 2027 have until 2 August 2030 (art. 1 point (39)(a)). Since 2 August 2026, providers, the developers that place AI systems on the market, must design chatbots so that people know they are dealing with AI (art. 50(1)).^2,4

The working group of France’s Cour de cassation, its highest court for civil and criminal cases, expects most justice uses to fall outside the high-risk category, as the exceptions are “numerous and relatively broad” (report, p. 2). The charter of the Conseil d’État, France’s highest administrative court, excludes high-risk AI (p. 5). The Dutch judiciary keeps AI out of high-risk processes for now.^21,22,28

What has changed for consumer ADR and online dispute resolution in the EU?

The EU closed its online dispute resolution (ODR) platform in 2025. Under national rules applying from 20 September 2028, EU consumer ADR bodies that use automated means in decision-making must tell parties in advance and let them ask for review by a natural person.^6,7

The platform drew two to three million visitors a year but passed only about 200 cases a year to ADR bodies (Regulation (EU) 2024/3228, recitals 3, 4). It stopped taking complaints on 20 March 2025, and its rules were repealed from 20 July 2025 (arts 1, 2(2)). The EU’s single digital gateway now points to the Commission’s list of consumer ADR entities (art. 4).⁷

Directive (EU) 2025/2647 of 16 December 2025 rewrites art. 5(2) of Directive 2013/11/EU, the consumer ADR directive (art. 1 point (4)(b)). Member States must adopt the national rules by 20 March 2028 (art. 5). Besides the notice and human review (points (d) and (e)), every consumer ADR body must keep a non-digital route alongside online filing (points (a) and (b)). The Commission also had to set up, by 20 April 2026, a digital tool that lists ADR bodies and offers them free machine translation (new art. 20(8), added by art. 1 point (10)(c)).⁶

Under the directive, decision-making covers whether to deal with a dispute and its outcome, not “purely administrative or technical tasks” (recital 21). The European Commission’s draft guidelines of 19 May 2026, not yet adopted, say these safeguards apply “in parallel with the AI Act” (para 433). Beyond consumer disputes, the draft lists equality bodies and professional disciplinary bodies among the ADR bodies that point 8(a) can reach (para 429). Our dispute resolution guide sets out when a scheme’s outcome has legal effects.^5,6

What does the GDPR say about automated decisions?

The EU’s General Data Protection Regulation (GDPR) gives people a right not to be subject to a decision based solely on automated processing that has legal or similarly significant effects (art. 22(1)). The right does not apply where the decision is needed for a contract, authorised by law with safeguards, or based on explicit consent (art. 22(2)). In the contract and consent cases, the controller (the organisation responsible for the processing) must at least let the person obtain human intervention, express their point of view and contest the decision (art. 22(3)).⁸

A Commission proposal of 19 November 2025 would replace art. 22(1) and (2) (COM(2025) 837, art. 3 point 7). On 26 October 2026 it was still awaiting a decision by the European Parliament’s committees.^61,62

In OQ v Land Hessen, C-634/21 (7 December 2023), the EU Court of Justice read “decision” broadly (paras 44 to 46) and called art. 22(1) “a prohibition in principle” (para 52). A credit agency’s automated score is itself a decision where a third party “draws strongly on” it (para 73).⁹

In CK v Magistrat der Stadt Wien, C-203/22 (27 February 2025), the Court read the right to “meaningful information about the logic involved” in automated decision-making (art. 15(1)(h)) as a right to an explanation of “the procedure and principles actually applied” (para 58). Handing over a complex mathematical formula, such as an algorithm, is not enough (para 59).¹⁰

The AI Act adds its own right. A person subject to a decision based on the output of a high-risk Annex III system, with legal or similarly significant effects, may ask the deployer for “clear and meaningful explanations” of the system’s role (art. 86(1)). That right applies only where EU law does not already provide one (art. 86(3)). Both judgments were credit-scoring cases, so how they apply to a neutral, such as an arbitrator or mediator, who relies on AI is [still open](#not-settled).^2,9,10

What has the Council of Europe set for AI in courts?

The Council of Europe has given courts guidance rather than binding law. Each CEPEJ text below keeps a human in control of the decision. Its AI treaty, CETS No. 225, will bind its parties but is not yet in force. Our guide to AI in law covers how AI itself is regulated.^11,12,14,20

• European Ethical Charter on the use of AI in judicial systems (adopted 3 and 4 December 2018): five principles, including “under user control”.
• Assessment tool (4 December 2023): the final decision “should remain a human-driven activity” (p. 2).
• Note on generative AI (12 February 2024): enter only data already “available in the public domain” (D.5), and check every answer and reference (D.6).
• AI Advisory Board reflections (31 January 2025): automating standardised first-instance decisions that can be appealed is raised as a question (para 32), but the right not to be subject to a fully automated decision “should be guaranteed” (para 34).^11,12,13,17

The CEPEJ’s guidelines on the use of generative AI for courts, adopted on 4 and 5 December 2025, go further. Their main points:¹⁴

• Pilot generative AI in payment-order procedures, small claims and labour claims for uncontroversial amounts, where all evidence is documentary (para 36).
• Keep “full human judicial intervention” where fundamental rights, minors or liberty are at stake (para 37).
• Set protocols that may require parties to disclose generative AI use (para 38), and make notice of that use a procedural obligation, with sanctions (para 39).
• Never delegate decision-making, legal reasoning or the assessment of evidence (para 55). AI outputs “shall never be binding” on judges (para 58).¹⁴

The Framework Convention on AI, CETS No. 225, opened for signature on 5 September 2024. It applies to AI used by public authorities and those acting for them, and each party must also address private actors’ AI and declare how (art. 3(1)). It requires remedies (art. 14), and each party must seek to ensure that people are told when they are interacting with AI (art. 15(2)). It enters into force about three months after five signatories, including at least three Council of Europe member states, agree to be bound (art. 30(3)). The EU approved it on 21 April 2026, and the EU is the only party listed on the Council of Europe’s page (capture of 16 October 2026).^18,19,20

What rules have European courts set for their own use of AI?

No court text we checked in Europe lets AI replace a judge. The texts differ on which tools are allowed and what data may go in. None requires lawyers to disclose AI use; Ireland and England and Wales say responsible use need not be mentioned, depending on context.^{22,23,24,25,30,31,37}

In Germany, court AI tools are state projects, and in the sources we checked we found no judiciary-wide guidance (how we research). England and Wales are outside the EU; our guide to arbitration in London covers them. Our legal tech guide covers the European bars’ guide for lawyers.^{34,35,36,37,39}

What have European arbitral and mediation institutions published on AI?

European arbitral and mediation institutions have published AI guidance rather than rules. The SCC Arbitration Institute, the Vienna International Arbitral Centre (VIAC), the Chartered Institute of Arbitrators (Ciarb) and the International Bar Association (IBA) Mediation Committee have AI texts, and the arbitration texts keep the decision with the arbitrator. The 2026 Rules of the International Chamber of Commerce (ICC) and the 2020 Rules of the London Court of International Arbitration (LCIA) say nothing about AI. Our arbitration guide asks whether AI can be an arbitrator under national laws.^{40,42,44,45,52,67}

What did arbitral institutions ask the European Commission to change?

On 23 July 2026, 16 arbitral institutions and bodies asked the Commission, in a joint submission to its AI Office, to revise the European Commission’s draft guidelines of 19 May 2026, not yet adopted. Where the high-risk label applies to arbitration, they want it to attach to AI used by or for the arbitral tribunal when it decides, not to the tools institutions use to administer cases. They also asked the Commission to revisit how the draft describes investment dispute bodies (s. 5).^5,53,55

The signatories included the ICC, the LCIA, the SCC, DIS, VIAC, Ciarb, the International Centre for Settlement of Investment Disputes (ICSID) and the American Arbitration Association’s International Centre for Dispute Resolution (AAA-ICDR). Their reasons:⁵³

• Institutions administer proceedings and “do not decide the merits” (s. 1). Listing them as ADR bodies (draft para 429) therefore “risks attaching the resulting obligations to a wrong actor” (s. 2).
• Draft paras 411 and 412 leave judicial administration outside point 8(a). The letter asks the same for arbitral institutions and for bodies that only administer mediation, expert determination or dispute boards, without power over the merits (s. 3).
• An investment dispute is typically seen as between an investor and a host State, so the draft’s phrase “between businesses” does not fit it (s. 4).⁵³

As of 26 October 2026 we found no adopted version. The Commission’s guidelines page, in its capture of 6 October 2026, still says feedback will go into the final version “before adoption by the Commission”. Our AI rules tracker has the joint letter.^54,63

Have European courts ruled on AI misuse or automated decisions?

The EU Court of Justice has ruled on automated decisions under the GDPR in two credit-scoring cases, [OQ v Land Hessen and CK v Magistrat der Stadt Wien](#automated-decisions). Courts in England and Wales have dealt with AI-invented citations. As of 26 October 2026, we found no European court ruling on AI misuse in an arbitration or a mediation.^9,10,58

• In Ayinde, [2025] EWHC 1383 (Admin), the High Court of England and Wales held that lawyers must check AI research against authoritative sources ([7], [8]). In the UK’s Upper Tribunal, immigration judicial review claim forms now need a statement of truth that each authority cited exists, can be found from its citation and supports the point made ([2026] UKUT 81 (IAC), headnote 3). See what courts do about AI-invented citations.
• Norway’s Supreme Court, in its guidance for counsel rather than a ruling, makes lawyers personally responsible for all information they use and warns that AI can hallucinate, inventing legal sources (point 19, 10 April 2025).
• In a disclosed trial, the Rotterdam district court in the Netherlands used AI as a writing aid for sentencing reasons in a criminal case. It said AI took no decisions and those involved were told once judgment was given (26 March 2025; ECLI:NL:RBROT:2025:3778 and 3779).
• The one annulment of an award over AI use we found is outside Europe: the Québec case.^{29,32,58,59,60}

What should courts, institutions and parties in Europe do now?

Courts, institutions and consumer ADR bodies in Europe can prepare now for two dates: 2 December 2027 under the AI Act and 20 September 2028 for consumer ADR.^4,5,6

• Map every tool against the point 8(a) examples in the European Commission’s draft guidelines of 19 May 2026, not yet adopted (printed pp. 135 to 139), and check the map again once the guidelines are final.
• Institutions: keep administrative tools apart from tools a tribunal uses to decide, and settle before 2 December 2027 if you are the provider or the deployer of each.
• Courts: before using any high-risk system, plan the fundamental rights impact assessment, cross-referring to the data protection impact assessment where it overlaps (art. 27(1); art. 27(4), as replaced by the Omnibus).
• Consumer ADR bodies: keep a non-digital route alongside online filing and, if you use automated means in decision-making, build advance notice and review by a natural person, before 20 September 2028 (new art. 5(2)(b), (d), (e)).
• Keep case files out of external AI tools (Conseil d’État charter p. 10; Norwegian framework s. 5).
• Parties and counsel: check every authority, expect some courts to require disclosure of generative AI use (CEPEJ guidelines paras 38, 39), and be ready to confirm that you checked AI research and citations (Irish guidelines point 7).
• Be ready to explain “the procedure and principles actually applied” in a decision based solely on automated processing (GDPR art. 22(1); C-203/22, para 58).^{2,4,5,6,10,14,22,30,31,53,58}

What is not settled on AI in dispute resolution in Europe?

Seven points were still open on 26 October 2026.⁵

• The final guidelines: the European Commission’s draft guidelines of 19 May 2026, not yet adopted, list arbitration institutions as possible ADR bodies (para 429), which the joint letter disputes. They also carry the court examples over to ADR without saying how they fit arbitration (para 434).
• GDPR art. 22: whether, by analogy with C-634/21, an AI recommendation counts as an automated decision when a neutral “draws strongly on” it.
• Existing court tools: how the draft’s example of drafting decisions in small claims cases applies to tools already in use, such as the Frankfurt air-passenger prototype.
• Spain: unlike the Supreme Court tool of 1 October 2026, the judicial AI the justice ministry described on 23 October 2026, developed with the General Council of the Judiciary, would be “capable of proposing decisions”, although the final decision “should always” be the judges’ own.
• National rules: how each Member State writes Directive (EU) 2025/2647 into its law, due by 20 March 2028 (art. 5).
• The treaty: when CETS No. 225 enters into force (art. 30(3)).
• No express ban: none of the practices banned in art. 5 of the AI Act, including those the Omnibus added (art. 1 point (7)), is fully automated adjudication. How far recital 61 limits it, by saying “the final decision-making must remain a human-driven activity”, is not settled.^{2,4,5,6,9,18,20,26,27,34,53}