AI in Arbitration and Litigation

AI in arbitration and litigation: the short answer

AI in arbitration and in German court proceedings is no longer a future topic. Counsel use generative AI to review documents, summarise files and draft submissions; tribunals and courts experiment with tools to structure mass proceedings. No German statute prohibits these uses, but several legal frameworks limit them: lawyers' professional secrecy and outsourcing rules, data protection law, the EU Artificial Intelligence Act (AI Act), the parties' right to be heard and the personal duty of arbitrators and judges to decide. Institutional guidelines — notably from the Chartered Institute of Arbitrators (CIArb) and the Silicon Valley Arbitration & Mediation Center (SVAMC) — now set practical standards on disclosure and delegation.

For companies, the questions are concrete: may our lawyers put our documents into an AI tool, must the tribunal or the other side be told, can an award or judgment be challenged because AI was used, and how should arbitration clauses and procedural orders address the issue? This note answers those questions from a German and EU perspective. It focuses on commercial disputes: civil litigation before the regional courts and the new Commercial Courts, and arbitration seated in Germany under the ZPO and institutional rules such as those of the German Arbitration Institute (DIS) and the ICC.

How the AI Act applies to courts and tribunals

The AI Act (Regulation (EU) 2024/1689) classifies certain uses in the administration of justice as high-risk. Annex III, point 8(a) covers AI systems intended to be used by a judicial authority, or on its behalf, to assist 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 explains that ADR tools are high-risk when the outcome produces legal effects for the parties — which is the case for arbitration — and stresses that "the final decision-making must remain a human-driven activity". Purely ancillary administrative tasks, such as anonymising decisions or internal communication, are not high-risk.

The consequences fall mainly on the provider of such a system (risk management, data governance, documentation, human oversight, conformity assessment) and on the deployer — the court, institution or arbitrator using it — who must use it according to the instructions and ensure human oversight. Under the Digital Omnibus on AI (Regulation (EU) 2026/1744), published on 24 July 2026, the obligations for high-risk systems listed in Annex III now apply from 2 December 2027 rather than 2 August 2026. Our insight on the AI Act timeline sets out the revised dates.

General-purpose tools used by counsel for research or drafting are usually not high-risk in themselves. But counsel and parties must still observe the transparency rules of Article 50 AI Act where they apply, and the AI literacy duty of Article 4 for organisations deploying AI. See our page on the AI Act.

Using AI as counsel: confidentiality and professional rules

German lawyers are bound by professional secrecy (§ 43a(2) of the Federal Lawyers' Act, Bundesrechtsanwaltsordnung, BRAO) and by criminal liability for disclosing client secrets (§ 203 of the Criminal Code). They may give external service providers access to confidential information only to the extent necessary, after careful selection, and on the basis of a contract in text form that obliges the provider to secrecy and restricts its access (§ 43e BRAO). Additional requirements apply where the service is provided abroad. Cloud-based AI tools are service providers in this sense.

In practice, this means:

• using enterprise versions of AI tools with contractual confidentiality, no training on client data and EU or adequate-country data processing;
• a data processing agreement under Article 28 GDPR where personal data are involved;
• internal policies on which matters and data categories may be processed;
• verification of every AI-generated statement of fact, law and citation before it is filed.

The last point is not trivial. Courts in several jurisdictions have sanctioned lawyers for filing briefs citing non-existent cases generated by AI. German procedural law places responsibility for submissions squarely on counsel: the duty of truthfulness (§ 138(1) of the Code of Civil Procedure, ZPO) and professional duties do not change because a tool drafted the text.

AI in arbitration: tribunals, disclosure and delegation

Arbitrators decide personally. Under German arbitration law, as under most laws based on the UNCITRAL Model Law, the mandate is personal and the parties' right to be heard (§ 1042(1) ZPO) must be respected. An award based on reasoning that the parties had no opportunity to address — for example, facts or legal theories introduced by an AI tool that were never put to them — risks being set aside on public policy or due process grounds (§ 1059(2) ZPO).

Institutional soft law has developed quickly. The CIArb Guideline on the Use of AI in Arbitration (2025) addresses the use of AI by parties and arbitrators. It allows arbitrators to require disclosure of AI use where it may affect the evidence, the outcome or the integrity of the proceedings, treats any disclosure duty as continuing, and warns tribunals not to delegate decision-making tasks to AI. The SVAMC Guidelines take a similar approach for technology-related arbitrations. These documents are not binding unless incorporated, but tribunals increasingly refer to them in procedural orders.

Addressing AI in procedural orders

Practical solutions are best agreed at the first case management conference. A short paragraph in Procedural Order No. 1 can define permitted tools, require disclosure of AI use for evidence or expert reports, confirm that arbitrators will not use AI to draft reasoning on contested issues, and set confidentiality standards for any tool used. Doing so early avoids disputes about disclosure later, when an AI-assisted expert report or translation becomes contentious.

Drafting clauses on AI in arbitration

Companies negotiating new arbitration agreements can address AI in arbitration directly or leave it to the tribunal. A short clause might state that the tribunal shall not delegate any part of its decision-making to AI tools, that the parties will agree on disclosure rules for AI-generated evidence at the first case management conference, and that any tool used must meet agreed confidentiality standards. Over-detailed clauses risk becoming outdated as technology changes; referring to a recognised guideline "as amended from time to time" or reserving the issue for the procedural order is usually more robust.

Arbitrators' declarations of independence and availability are another lever. Parties can ask candidates during the selection process how they use AI tools and whether they would accept limits. Several institutions are reviewing their rules and guidance, and the topic of AI in arbitration is likely to appear in future revisions of institutional rules, including those used for German-seated cases. Until then, party agreement and procedural orders remain the main instruments.

AI in German state courts

German courts use AI mainly to manage mass litigation and administrative workload; several state judiciaries have piloted assistance systems that categorise submissions and extract data in high-volume case types such as diesel and air passenger claims. These tools support case management; judgments remain the judge's personal responsibility, as constitutional guarantees of the lawful judge and judicial independence require. Once the AI Act's high-risk rules apply, tools that assist judges in researching and applying the law will have to meet those requirements.

For parties, the most visible effect is speed and standardisation in mass proceedings. In complex commercial cases, AI matters more on the counsel side: technology-assisted review of large document sets, chronologies, translation and analysis of expert reports. Where documents must be produced in court, parties should be able to explain how a review was conducted. See our commercial litigation and international arbitration pages for the procedural framework.

AI-generated evidence and expert reports

AI-generated or AI-processed material is increasingly offered as evidence: transcripts, translations, data analyses, reconstructed images, and expert reports prepared with AI assistance. German courts assess evidence freely (§ 286 ZPO), and arbitral tribunals determine admissibility and weight (§ 1042(4) ZPO). The opposing party can challenge reliability, and the tribunal or court can order disclosure of the methodology. Court-appointed experts may not delegate their assignment and must name assistants (§ 407a(3) ZPO); using AI does not change that responsibility. Our insight on expert evidence covers the expert's duties in more detail.

Deepfakes raise a separate issue. Where the authenticity of audio, video or images is disputed, parties should secure original files and metadata early and consider forensic analysis. The AI Act's labelling duties for deep fakes apply to deployers in many contexts, but they do not resolve evidentiary disputes in individual cases; see our insight on labelling AI-generated content and deepfakes.

Recommendations for companies

1. Policy: adopt an AI use policy for disputes covering in-house teams and external counsel, aligned with your AI literacy measures.
2. Clauses: consider whether new arbitration clauses should refer to an AI guideline or reserve the issue for the first procedural order.
3. Confidentiality: check that tools used on your matters meet contractual confidentiality and data protection standards.
4. Disclosure: decide in each case whether to propose disclosure rules for AI-generated evidence and expert work.
5. Verification: require human verification of all factual and legal statements before filing.

How TRW & Co helps with AI in arbitration and litigation

We advise companies on the use of AI in arbitration, litigation and internal investigations: drafting AI policies and outside counsel guidelines, negotiating AI provisions in procedural orders, addressing the reliability of AI-generated evidence and expert reports, and assessing whether AI use by a tribunal or the other side gives rise to challenges. We also advise providers of legal and dispute-resolution AI tools on their obligations under the AI Act. Where we use technology on client matters, we do so under the confidentiality framework of the BRAO and the GDPR and with human review of every output. To discuss AI in a pending or planned dispute, contact us.