Criteria for crossing the product liability threshold for a substantial modification of medical artificial intelligence

The revised Product Liability Directive (rPLD) introduces a key change in the liability landscape for healthcare professionals using AI. While existing research often focuses on professional or regulatory thresholds, this paper highlights a critical and under-researched milestone: the point at which a healthcare professional’s interaction with an AI system constitutes a significant modification, transforming them from a user into a de facto producer under the rPLD’s strict liability regime.
Drawing on the authors’ previous work on threshold-based liability frameworks, this paper examines how the rPLD redefines “product” to include AI software and systems and extends liability to any natural or legal person who significantly modifies a product beyond the control of the producer. In clinical practice, such modifications can occur through actions such as ignoring safety parameters, integrating unauthorized components, or using certified AI outside its intended purpose. When these actions change the fundamental safety values of an AI-enabled medical device, the healthcare professional (the institution in which they work) crosses the threshold of product liability, triggering strict liability for any resulting damage. The analysis places this threshold within the broader EU regulatory ecosystem to demonstrate how liability becomes distributed and context-dependent. The paper argues that rPLD not only complements existing fault-based regimes, but also creates a distinct, strict liability pathway that redistributes risk to the party with the greatest degree of control over the final safety configuration of the AI system.