Background and Context of the AI Liability Directive Proposal
The European Commission introduced the AI Liability Directive proposal in 2022 as part of a broader strategy to address legal and operational challenges posed by artificial intelligence systems. The proposal aimed to adapt non-contractual civil liability rules to the unique characteristics of AI, which include complexity, autonomy, and opacity—often referred to as the black box effect. This initiative was driven by concerns that existing fault-based liability frameworks were inadequate for handling claims related to AI-caused harm. A 2020 representative survey highlighted liability as one of the top three barriers to AI adoption among European companies, with 43% of businesses citing it as a significant obstacle to implementation.
The proposal was framed within the Commission’s Political Guidelines, which emphasized fostering trust and excellence in AI while addressing associated risks. It followed the European Parliament’s resolution of 20 October 2020, which called for a civil liability regime for AI under Article 114 of the Treaty on the Functioning of the EU (TFEU). The Commission’s Report on AI Liability, accompanying the White Paper on AI, further identified the challenges AI posed to existing liability rules, particularly the difficulty victims face in proving fault and causality due to the opaque nature of AI systems.
The proposal was designed to reduce legal uncertainty for businesses developing or using AI, particularly small and medium-sized enterprises (SMEs), which lack in-house legal expertise or capital reserves. Without harmonized EU rules, Member States were expected to introduce fragmented national adaptations, exacerbating legal uncertainty and increasing compliance costs for cross-border businesses. The open public consultation supporting the Impact Assessment confirmed these concerns, with respondents highlighting the black box effect as a major barrier to proving fault and causality in AI-related claims.
The directive was intended to complement other EU initiatives, including the AI Act, which focuses on safety and fundamental rights protections for high-risk AI systems. While the AI Act aims to mitigate risks before they materialize, the liability directive would address residual risks by ensuring victims of AI-caused harm receive equivalent protection to those affected by traditional products. This dual approach was meant to reinforce trust in AI by combining preventive safety measures with compensatory liability rules.
The proposal also aligned with the broader EU digital strategy, which prioritizes technology that serves societal needs. It was part of a package of measures to promote AI uptake, alongside revisions to product safety rules and sector-specific regulations for AI-enabled machinery and radio equipment. The directive was designed to avoid conflicts with other digital policies, such as the Digital Services Act (DSA), which focuses on intermediary liability for online platforms, and the Cyber Resilience Act, which targets cyber vulnerabilities in digital products.
Key Provisions of the Proposal
This measure was particularly important given the complexity and opacity of AI systems, which often make it difficult for victims to independently gather evidence.
Under this provision, courts could presume causality in certain cases, shifting the burden of proof to defendants to disprove a causal link between the AI system and the damage suffered. This approach was intended to streamline legal proceedings and reduce the financial and procedural barriers victims encounter when pursuing compensation. The proposal also sought to harmonize liability rules across the EU, preventing a patchwork of national adaptations that could create legal uncertainty for businesses operating across borders.
The directive was structured to complement the Product Liability Directive, which governs no-fault liability for defective products. While the Product Liability Directive primarily covers individual compensation for physical harm caused by defective products, the AI Liability Directive would address broader liability claims based on fault, covering any type of damage and victim. Together, these directives were intended to create a cohesive civil liability framework for AI, ensuring that victims of AI-caused harm—whether physical, financial, or reputational—could seek redress without prohibitive legal hurdles.
The proposal also included provisions to incentivize compliance with the AI Act’s safety requirements. Failure to adhere to high-risk AI system regulations under the AI Act would trigger alleviations in the burden of proof for liability claims, reinforcing the link between safety and accountability. The directive’s focus on transparency and accountability was intended to foster trust in AI systems while ensuring that victims of harm could access fair compensation.
Withdrawal and Current Status
The AI Liability Directive proposal was withdrawn in February 2025 due to a lack of agreement among EU institutions. Despite its withdrawal, the directive remains frequently cited in discussions on AI governance, particularly as its core objectives—addressing the black box effect, reducing legal uncertainty, and harmonizing liability rules—have been partially incorporated into other regulatory frameworks. The most significant development in this regard is the 2024 revision of the Product Liability Directive, which now encompasses software and AI liability under its scope.
It marked an early attempt to create a specialized legal framework for AI-related harm, addressing gaps in existing liability rules that were increasingly evident as AI systems became more pervasive in society. The directive’s failure to progress highlights the challenges of balancing innovation with accountability, particularly in an evolving technological landscape where legal precedents are still being established.
The shift toward integrating AI liability into the Product Liability Directive reflects a pragmatic approach to regulatory harmonization. By extending the scope of the Product Liability Directive to include software and AI, the EU has taken a step toward consolidating liability rules under a single framework, reducing fragmentation and ensuring consistency across Member States. This approach aligns with the broader trend of adapting existing legal instruments to emerging technologies, rather than creating entirely new regulatory structures. However, it also raises questions about whether the Product Liability Directive’s no-fault liability model is sufficiently flexible to address the unique challenges posed by AI systems, such as their autonomous decision-making capabilities and the difficulty of attributing fault in complex AI-driven processes.
The withdrawal of the AI Liability Directive does not preclude future efforts to address AI-specific liability issues. As AI systems continue to evolve, new legal challenges will emerge, necessitating further adaptations to civil liability rules. The experience of the AI Liability Directive proposal serves as a valuable case study in the complexities of regulating AI, offering insights into the need for collaborative, cross-sectoral approaches to ensure that liability rules keep pace with technological advancements.
Required actions
- No obligations: the proposal would have introduced a rebuttable presumption of causality and disclosure of evidence for damage caused by AI systems