Defensive medicine in the era of artificial intelligence

Abstract

The introduction of AI systems in healthcare may fuel new forms of defen-sive medicine, with consequences for the allocation of civil liability. Starting from the tension between adherence to clinical guidelines (Law No. 24/2017), good medical practice, and the use of innovative yet opaque tools, it is necessary to highlight the risks of a purely defensive deployment of AI-enabled care pathways. One can envisage scenarios of a “new” defensive medicine: unnecessary additional tests; exclusive reli-ance on certified AI; over-documentation to justify compliance with, or departure from, algorithmic outputs; risk transfer onto insurers, with downstream cost increases for patients; and refusal to use AI, with a possible loss of benefit for the patient. From a liability perspective, a distinction should be drawn between AI-enabled devices op-erating under physician control — which preserve the ordinary assessment of profes-sional negligence — and highly autonomous, opaque systems, where disruption of the causal link argues for shifting the centre of gravity towards manufacturer and supply-chain liability, also in light of the 2024 Product Liability Directive and the AI Act’s risk-based regime for “high-risk” systems. As regards healthcare providers, the patient’s contractual claim remains, together with a possible graduated right of recourse against the physician and the device manufacturer; a non-detectable malfunction may exclude the physician’s direct liability. The balance between innovation and patient protection hinges, on the one hand, on promoting trustworthy and transparent AI, coupled with strict (objective) manufacturer liability in cases of “machine error”; and, on the other, on recognising a possible negligent omission where a physician unjustifi-ably refuses to use adequate AI.