July 10, 2026
Year of AI 2026 · Updated July 2026
SAUDI COMPUTE
The Kingdom's Compute Buildout, Tracked.
Sovereign AI Infrastructure · Capital Flows · Geopolitical Intelligence

EPRS BRI(2023)739342 EN · Table (p.8)

From EPRS_BRI(2023)739342_EN.pdf · page 8

47 rows × 2 cols pdfs_from_pdf_links

01
EPRS | European Parliamentary Research Service
Relation to national laws
The draft AI liability directive leaves some margin of appreciation and interpretation to Member
States' legal orders. The directive would lay down EU rules for presumption of causality but not
harmonise rules regarding which party has the burden of proof or which degree of certainty is
required as regards the standard of proof. This remains a Member State competence within national
laws. 25 Furthermore, the proposed directive follows a minimum harmonisation approach. This
would allow claimants to invoke more favourable rules under national law (e.g. reversals of the
burden of proof under national fault-based regimes or national no-fault liability), for instance in
cases of damage caused by AI systems.26
Review clause
The Commission will have to submit a report to the Parliament, the Council, and the Economic and
Social Committee, assessing the achievement of the AIliability directive five years after its
transposition (Article 5). The AI liability directive therefore proposes to leave the door open for
future legislative development. In particular, that review should examine whether there is a need
to create no-fault liability rules for claims against the operator combined with a mandatory
insurance for the operation of certain AI systems, as suggested by the European Parliament
resolution of 20 October 2020 on a civil liability regime for artificial intelligence (2020/2014(INL)).
Such a review should consider, in particular, risks involving damage to important legal values, such
as the life, health and property of unwitting third parties, through the operation of AI-enabled
products or services as well as the insurance market's development of appropriate solutions.
Advisory committees
The European Economic and Social Committee (EESC) adopted an opinion on the proposal on
25 January 2023. The European Committee of the Regions (CoR) has not adopted an opinion.
National parliaments
The subsidiarity deadline for national parliament was set at 28 November 2022. In its submission,
the German Bundesrat warns that the term 'AI system' in the AI liability directive is based on a
definition proposed in the draft AI act that would lead to an excessively broad scope of application.
Furthermore, the Bundesrat stresses that, under the current draft text, in the case of a claim for
damages concerning an AI system that is not a high-risk AI system, the presumption of causality
would only apply where the national court considers it 'excessively difficult' for the claimant to prove
the causal link (Article 4(5)). This requirement rests therefore on an imprecisely defined legal
concept.
Stakeholder views27
Some of the main contentious points raised so far are listed below.
Presumption of causality, burden of proof and coherence with the
proposed AI act
Consumer associations and civil society groups welcome the proposed new EU liability rules but
warn they contain blind spots. The main EU consumer protection association, BEUC, criticises the
proposal for placing the responsibility on consumers to prove the fault lies with the operator.
Considering how opaque and complex AI systems are, BEUC argues these conditions will de facto
make it impossible for consumers to use their right to compensation for damages. The Future of Life
Institute (FLI), an independent non-profit organisation, recommends a strict liability regime for high-
risk AI systems as well as for general purpose AI systems and proposes that all other AI systems fall
under a fault-based liability regime where the presumption offault lies on the operator.
Furthermore, FLI calls upon EU lawmakers to harmonise the immaterial damages and indirect harms
Source
EPRS_BRI(2023)739342_EN.pdf
Domain
pdfs_from_pdf_links
Type
pdf
Method
camelot_stream
Dimensions
47 × 2
Page
8