Presumption of defectiveness
A de facto reversal of the burden of proof introduced by the 2024 directive, triggered notably by the product's technical complexity.
Directive (EU) 2024/2853 on liability for defective products introduces rebuttable presumptions that lighten the victim's burden of proof. One of them applies where that proof is excessively difficult to produce because of the product's technical or scientific complexity. The mechanism shifts the burden without reversing it: the victim establishes the difficulty and the likelihood, and the manufacturer must then show the product was not defective. A second presumption applies where the defendant fails to produce relevant evidence in its possession that the court ordered it to disclose. Applied to a learning system, this construction turns the model's opacity against the party that designed it. Unexplainability used to serve as a technical argument against any reconstruction of the reasoning. It now becomes a legal liability, since it is precisely what makes proof excessively difficult and therefore triggers the presumption. The defense lies upstream, in training and validation documentation kept so it can be produced.
Textbook case: a maker of an embedded AI system must open the black box and produce its training documentation, or see defectiveness presumed.
Follow the research
One letter a week. First, the week's underwriting note, when one is published: a dated, argued decision, signed by Paul Bernardski. Then what the observatory published on cyber risk, political risk and insurance. No account, no tracking.
No spam, one-click unsubscribe, no data ever sold.