GERMANY Trends and Developments Contributed by: Tanja Pfitzner and Claudia Löschhorn, Pfitzner Legal (in collaboration with aginom)
In March 2025, the Chartered Institute of Arbitrators (CIArb) issued a particularly detailed document based on the principle that the use of AI does not alter the allocation of responsibility, ie, each participant remains as accountable as if the tool had not been used. EU AI Act Annex III, point 8 (a) of the EU AI Act takes the same approach. It classifies AI systems as “high-risk” if they are intended to be used by or on behalf of, a judicial authority to assist in researching and interpreting facts and the law and in applying the law to a concrete set of facts. It extends the same classification to systems intended for use in a similar way in alternative dispute resolution, where the outcome has legal effects for the parties. The Commission’s draft guidelines of 19 May 2026 remove any doubt that the scope of the AI Act also encompasses arbitration. They define ADR as adjudicating a dispute out of court with the assis - tance of an impartial ADR body and list arbitration first among the examples (Annex III Guidelines, Para - graph 426). Recital 61 of the AI Act explains explains this classification by referring to the risks of potential biases, errors and opacity. It also states that while AI may support decision-making, the final decision must remain a human-driven activity. The Digital Omnibus on AI entered into force on 27 July 2026 and rewrote Article 113 of the AI Act: The high- risk obligations for stand-alone Annex III systems now apply from 2 December 2027 (delayed from 2 August 2026). Nevertheless, there are other obligations under the AI Act, such as the AI literacy duty in Article 4, which already applies since 2 February 2025 and the Article 50 transparency obligations in effect from 2 August 2026 that have not been deferred. It should also be noted that the Commission’s draft guidelines on the classification of high-risk AI systems state that human involvement cannot alter a system’s intended purpose and therefore has no bearing on whether it falls within the high-risk category. Consequently, the harness reduces risk and makes reliance defensible without changing the classification. Where a system genuinely performs only a narrow procedural task or merely improves a completed human analysis, Article 6 (3) may take it outside the regime, which is why the line between document handling and evaluative use
has to be drawn deliberately, bearing in mind that Arti - cle 6 (3) is subject to narrow interpretation. In terms of the allocation of obligations, the AI Act distinguishes between providers and deployers. In principle, an arbitrator who uses a commercial tool is a deployer (unless the system is substantially modi - fied or the arbitrator puts their name on it). Its duties under Article 26 (human oversight, relevant input data, monitoring and log-keeping) are quite similar to those of the harness proposed above. What remains open is exactly who the AI Act refers to: the arbitral institution or the arbitral tribunal. To clarify this, a group of inter - national arbitral institutions, coordinated through the LCIA, in a joint submission to the consultation asked the Commission to confirm that the classification attaches to systems used by or on behalf of an arbi - tral tribunal in its adjudicative function, rather than to systems used by institutions for purely administrative or organisational purposes. The consultation closed on 23 July 2026. Final guidelines are expected by the end of the year. They will remain non-binding and only the Court of Justice can settle the meaning of the Act. A comparison of these standards, guidelines and leg - islation reveals that global developments are remark - ably consistent in the point that the tool may assist, but the human has to decide. A Working Checklist for Arbitrators Regarding Bias What does all this mean on a Tuesday morning, with a real case file open? The following is a checklist, not a code, and it deals with bias only. Other risks of AI- use, eg, data protection and privacy, confidentiality, IP, data governance and data security, are not con - templated in this checklist. It should also be kept in mind that technological and legal developments are currently subject to constant change. The checklist does not reproduce any institution’s guidance and not every case requires all of the steps listed. Early raising Any proposed evaluative use of AI in arbitration should be placed on the agenda of the first case management conference, the parties’ views should be sought and a record should be made of what was agreed upon or decided. Where appropriate, it should be identi - fied which uses are permitted, restricted or prohib -
398 CHAMBERS.COM
Powered by FlippingBook