Liability for Harm Caused Through the Use of Artifi cial Intelligence: Limits of European Harmonisation

In 2022, the European Commission introduced a draft directive on liability for harm caused through the use of artificial intelligence systems. The hallmark of the proposal lies in its decision to retain a fault-based liability regime, even when harm arises in connection with the deployment of AI technologies. The draft further includes several procedural mechanisms ostensibly aimed at alleviating the burden on claimants — most notably, provisions on disclosure of evidence and a rebuttable presumption of causality. This article offers a critical analysis of these proposals in light of the broader European doctrinal discourse. The authors argue that the draft directive fails to adequately address the legal and technological complexity of modern AI, particularly large language models such as ChatGPT and GPT-4. The classification scheme for AI systems adopted by the draft appears arbitrary and may further complicate the claimant’s position. Additionally, the proposed disclosure mechanism overlooks the so-called „many hands“ problem inherent in distributed AI architectures. The preconditions for invoking the presumption of causality are so narrowly framed as to render the presumption effectively inapplicable in practice.