AI Liability Directive (withdrawn)

Proposal of 28 September 2022 for a Directive of the European Parliament and of the Council on adapting non-contractual civil liability rules to artificial intelligence

Background and Scope

The proposed AI Liability Directive was put forward by the European Commission as part of a broader package to support the responsible deployment of AI across the EU. Liability had been identified as one of the three biggest barriers to AI adoption among European companies, cited as the most relevant external obstacle for businesses yet to deploy the technology.

Existing national liability rules, particularly those based on fault, were not suited to handle claims for damages caused by AI-based products and services. The characteristics of AI, including its complexity, autonomy and lack of transparency (the so-called “black box” effect), could make it difficult or disproportionately costly for victims to identify the responsible party and meet the requirements for a successful claim.

The proposal therefore aimed primarily to make it easier to pursue non-contractual liability claims for damages caused by AI, by ensuring that victims of AI-caused harm receive the same level of protection as those harmed by products in general, while reducing legal uncertainty for businesses developing or deploying AI systems. Without EU action, Member States were expected to adapt their national liability rules individually, leading to further fragmentation and increased costs for businesses operating across the EU.

The proposal formed part of a package together with a revised Product Liability Directive. The two instruments covered different types of liability, as the Product Liability Directive addressed strict liability for defective products, while the AI Liability Directive covered fault-based national claims for any type of damage and any type of victim.

Key Obligations

• Disclosure of evidence: Providers, users and other operators of high-risk AI systems may be ordered by national courts to disclose relevant evidence where a claimant has presented sufficient facts to support the plausibility of a claim for damages caused by the AI system.
• Rebuttable presumption of non-compliance: Where a defendant fails to comply with a court order to disclose or preserve evidence, national courts shall presume that the defendant has failed to comply with the duty of care to which the requested evidence relates. The defendant may rebut this presumption.
• Rebuttable presumption of causality: National courts shall presume a causal link between a defendant’s fault and the output of an AI system where the defendant has breached a duty of care intended to protect against the damage suffered, it is reasonably likely that the fault influenced the AI output, and the claimant has shown that the output caused the damage. The defendant may rebut this presumption.