GERMANY Trends and Developments Contributed by: Tanja Pfitzner and Claudia Löschhorn, Pfitzner Legal (in collaboration with aginom)
Pfitzner Legal Kettenhofweg 98 60325 Frankfurt am Main Germany
Tel: +49 69 87 000 29 00 Fax: +49 69 87 000 29 99 Email: pfitzner@pfitznerlegal.com Web: www.pfitzner.legal aginom
Unter den Nussbäumen 16 65779 Kelkheim (Taunus) Germany Tel: +49 162 15 81818 Email: cl@aginom.com Web: www.aginom.com
Dealing with Bias in AI-Assisted Arbitration On 22 April 2026, the Québec Cour Supérieure issued what appears to be the world’s first decision in which a court set aside an arbitral award on the grounds of the use of generative AI (ARIHQ v Santé Québec, 2026 QCCS 1360). The Canadian court, which was hear - ing a motion to set aside a domestic arbitral award, based its decision on Article 646 (3) of the Québec Code de procédure civil, which – drafted on the basis of Article 34 (2)(a)(iv) UNCITRAL Model Law, closely mirroring Article V(1)(d) of the New York Convention – allows an annulment where the applicable procedure was not respected. The court held that when parties agree to arbitrate, an arbitrator is selected precisely because of his or her personal expertise and judg - ment and that the parties are entitled to have their dispute decided and the reasons formulated, by that arbitrator. If the substantive reasoning is delegated to an AI tool without the result having been sufficiently reviewed, this violates the principle of “delegatus non potest delegare” and also breaches the confidential - ity of deliberations. While the court emphasised that not every use of AI would, per se, lead to annulment, it also pointed out five risks associated with its use in decision-making processes, as outlined below. • Hallucinations: AI may invent sources that appear deceptively real.
• Lack of human discretion: AI may not adequately account for human values and the specific circum - stances of a case. • Bias: AI operates on a statistical basis. It processes text, images and conversations and reproduces what appears most likely based on contextual elements. Thus, AI systems can be unintentionally programmed to exhibit biases due to program - mers’ prejudices or learn to be biased from the data used to train them. These risks are exacer - bated by the fact that the companies that create the algorithms hold the associated trade secrets and are unwilling to reveal the source code or even discuss how the algorithm works (this opacity is sometimes referred to as the “black box” phenom - enon). • Confidentiality may be compromised by the use of AI. • Lack of trust: Decisions made by machines may undermine public confidence in the justice system. The court held that it is crucial that responsibility for the reasoning remains with the decision-maker. Whether the use of AI leads to annulment depends on the individual case. In the case under review, the violation was deemed significant because all legal ref - erences the arbitral tribunal relied upon as the legal basis for the award (scholarly articles, court decisions
394 CHAMBERS.COM
Powered by FlippingBook