The Proof Problem: Evidence, Opacity, and Presumptions

Directive (EU) 2024/2853 keeps the claimant’s basic burden of proof even while it extends no-fault product liability to software and AI systems. This chapter focuses on the bridge between a plausible claim and a provable one: when can a claimant request evidence held by an operator, when can a court presume defect or causation, and how does technical complexity alter the analysis? The result is a calibrated model in which opacity can justify procedural relief without turning every unexplained AI output into automatic liability.

No fault still requires a causal chain

Under Article 10(1), a claimant must prove the product’s defectiveness, the damage suffered, and the causal link between the two. The operator does not need to have acted negligently for liability to arise, but the claimant still has to connect the product to the covered harm. A defective product that did not cause the loss is not enough. A real loss with no defect is not enough. No-fault liability removes one inquiry—the operator’s state of care—but it does not remove the structure of the claim.

That structure can be hard to satisfy for a digital product. A claimant may know that an AI-enabled appliance behaved dangerously, but not which model version produced the output, what update changed the behaviour, what data the system received, or which operator controlled the relevant service. The operator may hold the technical records, design history, testing material, service configuration, and information needed to explain the causal path. The imbalance is not proof that the operator is wrong. It is a reason the ordinary assumption that both sides can gather the same evidence may fail.

Article 9 responds to that imbalance with a targeted disclosure mechanism. A claimant who has presented facts and evidence sufficient to support the plausibility of a compensation claim can ask the national court to require the defendant to disclose relevant evidence at the defendant’s disposal. The threshold matters. The Directive does not authorise a claimant to demand every document from a company merely by naming a product. There must be enough initial material to make the claim plausible and to show why the requested evidence is relevant.

The mechanism is reciprocal. A defendant that presents facts and evidence sufficient to demonstrate a need for evidence to counter the claim can ask the claimant to disclose relevant evidence in the claimant’s possession. This recognises that an operator may need information about how the product was used, what happened immediately before the event, or what damage was actually suffered. Disclosure is not designed as a one-way penalty for being the better-informed party.

Disclosure is broad enough to help, but limited enough to remain fair

Article 9 requires disclosure to be necessary and proportionate. A court must consider the legitimate interests of all parties, including third parties, especially where confidentiality or trade secrets are involved. The proportionality requirement has two functions. It prevents the request from becoming an indiscriminate search for anything embarrassing or potentially useful, and it keeps the evidential remedy connected to the defect, damage, or causal link that the claimant must establish.

The Directive’s recitals recognise that digital evidence may need to be compiled or classified by the defendant rather than simply handed over as one existing file.