• Citer

Looking through the crack in the black box: A comparative case law benchmark for auditing AI-Powered Trade Mark search engines

par Cabay, Julien ;Vandamme, Thomas ;Debeir, Olivier
Référence Computer law & security review, 59, 106167
Publication Publié, 2025-11

Article révisé par les pairs

• Accès en ligne
• Détails
• Contenu
• Statistiques

Résumé :

For the past few years, Intellectual Property (IP) Offices have provided their users the possibility to carry out searches in the Trade Mark (TM) public registries through image-search tools, powered by Artificial Intelligence (AI) technologies. Such tools allegedly alleviate the burden to identify similar figurative trade marks (TM), which is a crucial yet cumbersome task for TM proprietors, TM applicants and IP Offices. Amongst others, the European Union Intellectual Property Office (EUIPO) and the Benelux Office for Intellectual Property (BOIP) provide access to such tools, respectively developed in-house and by a private company. Yet, the inner functionings of those systems are unknown and their performances difficult to assess, which in turn raises many concerns, especially in light of the legal certainty rationale underlying the registration requirement of TM law. To address those concerns, we designed an experiment to benchmark and audit those tools. Using the case law from the EUIPO and the BOIP on opposition to TM registration, we evaluated the capacity of those tools to identify similarities between signs that possibly amount to a likelihood of confusion (LoC), the main trigger of TM law. Our findings show that the performances of those tools are poor, and that the black-box auditing is highly contingent and possibly elusive for many AI technologies used in the legal field. This suggests that black-box auditing is not suitable for Legal AIs, which should be subject to enhanced transparency obligations, possibly pursuant to the AI Act interpreted broadly.