AI Liability Directive

On 28 September 2022, the Commission published the Proposal for a Directive of the European Parliament and of the Council on the adaptation of the rules on non-contractual civil liability to artificial intelligence (AI Liability Directive). 

As it is still a legislative proposal, it is not possible to know at this time when it will be published or enter into force.

Objective

The AI Liability Directive will aim to establish rules that will make it possible to determine who will be liable for damages arising from an AI system in those cases where this issue had not been previously foreseen in a contract. This will create a uniform liability regime with the aim of improving the functioning of the European internal market.

This regulation could be useful, for example, in case the AI system used by a city’s waste management company to plan garbage collection was not precise enough with the regularity in which the containers in a street must be emptied, causing inconvenience to the people who live in it.  

Relationship with the AI Act

The directive will work in a complementary manner to the AI Act, since the AI Liability Directive refers to the obligations contained in the AI Act to identify the subjects on whom these rules are imposed as presumed guilty of the damages complained of. 

It is worth mentioning that while the AI Act is a European regulation and is therefore applicable from the moment it was published in the Official Journal of the EU, the AI Liability Directive is a European directive. Consequently, even when the OJEU has been approved and published, its content will not be enforceable in a territory until the state has created its own rule by transposing its content.

Content

Article 4(1) of the Directive provides that the defendant shall be presumed to be guilty where the following conditions are met: (1) the person who brought the action proves, or the court assumes, that the defendant, or someone under his responsibility, has breached a duty of care established by law (especially the requirements of the AI Act); (2) it is reasonably likely that this failure has influenced the results of the AI system, or its lack of results; and (3) the claimant must demonstrate that the damage was caused by the results produced or not produced by the AI system. 

The other paragraphs of Article 4 set out specific criteria for determining when the listed requirements are met, including specifications in relation to high-risk AI systems. 

Legislative background

Below you will find some summary tables with the legislative advances that have been carried out so far, as well as the links to access the different texts adopted.

Texts adopted by: