EU Proposes Sweeping AI Liability Rules to Build Trust and Foster Innovation
Brussels, Belgium – The European Commission today, January 20, 2025, unveiled a significant new legal proposal in Brussels aimed at modernizing and clarifying liability rules specifically for Artificial Intelligence (AI) systems across the European Union. This landmark framework represents a crucial step in the EU’s broader strategy to promote the uptake of trustworthy AI, ensuring that victims of harm caused by AI systems can obtain compensation while also providing legal certainty for businesses and developers.
The complexity and opacity inherent in many AI systems pose unique challenges to traditional liability rules, which are often predicated on concepts of fault and causality that can be difficult to prove when damage is caused by autonomous or semi-autonomous AI operations. Existing national liability laws within the 27 EU member states were not designed with these specific characteristics of AI in mind, potentially leaving individuals and businesses unable to seek effective redress when they suffer harm, be it personal injury, property damage, or even non-material loss, as a result of an AI system’s behaviour. This uncertainty can not only undermine public trust but also hinder the responsible development and deployment of AI technologies.
Recognizing this gap, the European Commission’s new legal proposal seeks to address these challenges head-on. It introduces targeted rules designed to make it easier for victims to obtain compensation when damages are caused by high-risk AI systems. The framework aligns closely with the principles and definitions established in the proposed EU AI Act, focusing particularly on those AI applications deemed to pose a higher potential risk to fundamental rights and safety.
One of the most significant innovations within the proposal is the introduction of a rebuttable presumption of causality for damages caused by high-risk AI applications. This mechanism is designed to tackle the specific difficulty claimants often face in demonstrating a direct causal link between the AI system’s operation and the damage suffered. Proving such a link typically requires access to technical information about the AI system, its algorithms, training data, and internal processes – information that is almost exclusively held by the developer or user of the AI system.
Under the proposed framework, if certain conditions are met, a court can presume that the AI system caused the damage. These conditions might include instances where the provider or user of a high-risk AI system has failed to comply with relevant obligations under the AI Act, and this non-compliance is plausibly linked to the damage that occurred. By establishing this rebuttable presumption, the framework effectively shifts the burden of proof onto the defendant – either the developer or the user of the high-risk AI system – to demonstrate that their AI system did not cause the damage.
This shift in the burden of proof is a pivotal element of the proposal. It acknowledges the significant information asymmetry that exists between the claimant and the party in control of the AI system. It aims to level the playing field, making it substantially easier for victims to navigate the legal process and obtain fair compensation. Instead of the victim having to undertake the often-impossible task of forensically proving how the AI system’s intricate workings led to the harm, the party with access to the technical details is now responsible for rebutting the presumption of causality. This does not create strict liability but rather adjusts the procedural rules for proving causation in complex AI-related cases.
The proposal also aims to provide greater clarity on the concept of ‘fault’ in the context of AI, particularly regarding damage caused by errors or failures of the system. It clarifies that existing product liability rules, which cover damages caused by defective products regardless of fault, will continue to apply to AI systems when they are incorporated into products.
The overarching objectives of this new liability framework are multifaceted. Firstly, it seeks to enhance consumer protection and the protection of citizens more broadly. By providing a clear and effective path to compensation for damages caused by AI, the proposal aims to ensure that individuals are not left vulnerable when interacting with increasingly prevalent AI technologies. This is seen as essential for building fundamental public trust in AI.
Secondly, the framework is intended to build trust in AI across the 27 EU member states. Legal certainty regarding liability is a cornerstone of trust for both users and potential victims. Knowing that mechanisms are in place to address potential harm encourages wider acceptance and adoption of AI technologies.
Thirdly, by clarifying the rules and reducing legal uncertainty, the proposal aims to facilitate responsible innovation. Developers and businesses operating in the AI space will have a clearer understanding of their potential liabilities, allowing them to manage risks more effectively and invest in developing safe and compliant AI systems. A harmonized approach across the EU will also reduce fragmentation and administrative burdens for companies operating cross-border.
This proposal is a complementary piece to the comprehensive AI Act, which lays down rules for placing AI systems on the market and using them. While the AI Act focuses on preventing harm by ensuring AI systems are safe and compliant from the outset, the liability framework addresses what happens when, despite preventative measures, harm does occur.
The announcement of this significant proposal on January 20, 2025, marks the beginning of its journey through the European Union’s legislative process. The proposed legal framework now moves to the European Parliament and the Council for detailed review, debate, and potential amendments. The legislative bodies will examine the text, gather input from stakeholders, and work towards reaching a final agreement. This process is expected to take some time before the framework is formally adopted and enters into force, shaping the future of AI liability across the Union. Its successful implementation is seen as vital for harnessing the benefits of AI while effectively managing its risks.
