Impending Revisions to the AI Liability Directive Following AI Act Implementation
Brussels, Belgium - Following the formal introduction of the EU Artificial Intelligence Act (AI Act), European Union (EU) institutions have shifted their focus towards refining the proposed directive on adapting non-contractual civil liability rules for artificial intelligence (AI), commonly referred to as the AI Liability Directive. While the European Parliament and the Council provisionally agreed on the directive's text in December 2023, recent developments prompted by a complementary impact assessment by the European Parliamentary Research Service (EPRS) indicate that substantial changes are on the horizon.
Overview of the AI Liability Directive
The AI Liability Directive is designed to create standardised rules for fault-based claims related to AI systems, addressing situations such as negligence. Key aspects of this directive include stipulations for the disclosure of evidence concerning high-risk AI systems and a framework for the burden of proof. Notably, it introduces a rebuttable presumption of causation in cases where a fault by the AI system's provider or deployer leads to damages.
Potential Changes to the Directive
In July, an amended version of the AI Liability Directive proposal was leaked, aligning its terminology with the recently enacted AI Act. Over the summer of 2024, the EPRS conducted an assessment to determine the necessity and scope of the AI Liability Directive in light of the forthcoming Product Liability Directive, which addresses strict liability issues for AI systems and is expected to be adopted in late 2024, coming into effect by autumn 2026.
The complementary impact assessment, published on 19 September, asserts the necessity of the AI Liability Directive but recommends several significant changes to its scope and application:
Transition to Regulation: The EPRS suggests converting the directive into a regulation directly applicable in all member states to eliminate discrepancies between national AI liability frameworks, believed to hinder AI developers and consumers. This shift aligns with recent trends in product safety and market regulation laws in the EU.
Extension to Non-AI Software: The assessment proposes broadening the directive's scope to include non-AI software, matching the proposed Product Liability Directive's inclusive approach. Additionally, the types of compensable harm under the AI Liability Directive should expand to cover areas such as discrimination, personality and fundamental rights, professional property, pure economic loss, and sustainability impacts.
High-Risk AI Systems: Provisions in the directive relating to high-risk AI systems should also cover newly identified areas of concern and AI systems prohibited under the AI Act. This includes general-purpose AI systems, transportation-related AI applications, and various insurance applications beyond health and life insurance.
Causal Link and Non-Compliance: The directive should explicitly establish a causal link between the output of an AI system and resultant damages in cases of non-compliance with human oversight provisions outlined in the AI Act. Failures by AI providers and deployers in ensuring adequate supervision would be presumed to cause the harmful outputs.
Disclosure of Evidence: The directive should facilitate claimants seeking court orders for the disclosure of evidence necessary to support their claims, provided harm and AI involvement are demonstrated. However, competitors of the defendant would be excluded from this provision to prevent vexatious litigation and protect trade secrets.
Joint Liability: The assessment recommends introducing joint liability along the AI value chain, with several proposed methods for fair liability distribution, including presumed equal liability among involved actors, exemptions for SMEs, and prohibitions against contractual clauses that limit downstream recourse rights.
The complementary impact assessment also advises a more detailed future evaluation of incorporating strict liability into the AI Liability Directive, noting various pros and cons.
Next Steps
In October, the European Parliament’s Legal Affairs Committee (JURI) will deliberate on whether to endorse the impact assessment's recommendation to transition from a directive to proposing an AI liability regulation. Although JURI is not mandated to follow the assessment's findings, these will significantly influence the political decision-making process.
Simultaneously, the Council has solicited feedback from member states on specific aspects, including claimant measures for identifying liable parties and the conditions for rebuttable presumption of causation. Member states have until 11 October to respond.
As the EU continues refining its approach to AI liability, stakeholders in the fields of AI development, data protection, and consumer law await further guidance on these evolving regulations.
Source: Noah Wire Services