Lugano, May 7, 2026
Author: Lawyer Enrico Germano
The growing use of artificial intelligence systems in decision-making processes presents Swiss law with a structural challenge: how to assign liability for damages generated by decisions that are no longer fully attributable to direct human behavior.
From automated personnel selection to diagnostic systems in healthcare, to algorithms used in the financial sector, AI no longer simply supports human action, but replaces it in significant segments.
In this context, the Swiss regulatory framework appears, at first glance, flexible enough to absorb these innovations. The general rules on civil liability continue to offer formally applicable instruments. However, this apparent adequacy conceals a deeper criticality: these rules were designed for a linear causal paradigm, based on the attribution of the harmful event to identifiable human conduct.
Comparison with the evolution of European law, particularly the AI Act, further highlights the risk of progressive regulatory misalignment. Without targeted intervention, Swiss law risks not only being ineffective in protecting those harmed but also creating legal uncertainty for economic operators.
It should be noted that in Switzerland there is no law (at least not yet) equivalent to the European Union’s AI Act, which, we recall, entered into force on August 1, 2024, but which envisages gradual implementation, with full application of most of the provisions expected by August 2026 (and which could be the subject of a specific article on this complex topic).
Switzerland has chosen a more gradual and less centralized approach: there is no single law dedicated to artificial intelligence, and in this case, existing laws, adapted to specific cases, apply, for example, data protection (Federal Act on Data Protection-FADP), civil liability, or criminal law.
- The Limits of Traditional Civil Liability
Swiss civil liability law is based on established categories, particularly liability for tort and specific cases of strict liability. These tools, however, presuppose the ability to reconstruct a causal link between attributable conduct and the damage that occurred.
In the case of artificial intelligence systems, this reconstruction faces significant obstacles.
First, the opaque nature of algorithms—often described as “black boxes”—makes it extremely difficult to demonstrate how a given decision was generated. This directly impacts the proof of fault and causality, central elements in the traditional system.
Second, the decision-making autonomy of more advanced systems weakens the connection between human behavior and harmful events. AI’s action is no longer simply the execution of predetermined instructions, but the result of learning and adaptation processes.
Finally, the complexity of the technological supply chain—which involves developers, data providers, integrators, and users—leads to a fragmentation of liability that is difficult to manage using traditional frameworks.
This creates a systemic tension: while existing regulations remain formally applicable, they are often ineffective in terms of evidence, with the real risk of leaving the injured party without effective protection.
This is a highly topical issue. A recent study by the Lucerne University of Applied Sciences (HSLU) of approximately 400 company executives in Switzerland, conducted between October 2025 and March 2026, found that over 60% of companies consider AI-related transformation to be too slow and ineffective (Newspaper La Regione, April 23, 2026).
In the next article, we will explore the aspects related to corporate compliance and de facto regulation, as well as the need to introduce a specific liability regime for AI in Swiss law.
