EPRS BRI(2023)739342 EN · Page 7

Artificial intelligence liability directive 
7 
 Third, the claimant has demonstrated that the output produced by the AI system or the AI 
system's inability to produce an output gave rise to the damage (Article 1(c)).  
By reducing the burden of proof, the AI liability directive intends to make it easier for people alleging 
injury from AI to succeed in bringing claims, given the complexity of the AI environment (i.e. 'black 
box').21 In practice, the new rule means that if a victim can show that someone was at fault for not 
complying with a certain obligation relevant to their harm, and that a causal link with the AI 
performance is reasonably likely, the court can presume that this non-compliance caused the 
damage.  
The defendant may, however, rebut this presumption of causality, for example by showing that 
its fault could not have caused the damage (recital 30).  
The proposed approach does not entail a reversal of the burden of proof, according to which the 
victim no longer bears the burden of proof and it is for the person liable to prove that the conditions 
of liability are not fulfilled. The Commission discards such a reversal of the burden of proof to avoid 
exposing providers, operators and users of AI systems to higher liability risks, which could hamper 
innovation in AI-enabled products and services.22 Under the proposed approach, the victim would, 
instead, still bear the burden of proof, but the presumption of causality would result in a targeted 
alleviation of the burden of proof regarding the question as to how or why an AI system reached a 
certain harmful output. This approach would relieve victims of the need to demonstrate the inner 
workings of the AI system at stake. 
Disclosure of evidence  
The large number of people potentially involved in the design, development, deployment and 
operation of high-risk AI systems, makes it very difficult for plaintiffs to identify the person 
potentially liable for damage caused and to prove the conditions for a claim for damages. To remedy 
this, the AI liability directive would give national courts the power to order disclosure of evidence 
about high-risk AI systems that are suspected of having caused damage (Article 3(1)). The new 
rules would help victims to access relevant evidence to identify the person that could be held liable, 
for instance, when damage is caused because an operator of drones delivering packages does not 
respect the instructions for use or because a provider does not follow requirements when using AI-
enabled recruitment services.23 Accordingly, companies responsible for high-risk AI systems would 
be required to disclose a range of information including specific documentation, information and 
logging requirements.24  
Requests are to be addressed to parties that bear obligations under the AI act, i.e. the provider of an 
AI system, a person who is subject to the provider's obligations laid down under the AI act or a user 
pursuant to the AI act. The claimant must present sufficient evidence to support the claim and make 
proportionate effort to obtain evidence from the defendant. The claimant could also request the 
disclosure of evidence from providers or users that are not defendants, but only in cases where all 
proportionate attempts made to gather the evidence from the defendant were unsuccessful. In 
addition, for the judicial means to be effective, the AI liability directive would also provide that a 
court may also order the preserv