Product Liability Directive (2024/2853)
Summary
Section titled “Summary”Directive (EU) 2024/2853 replaces the 1985/374 old Product Liability Directive; it entered into force on 2024-12-08, with member states required to transpose it into national law by 2026-12-09. Three points matter most for AI:
- Software / AI systems = products: expressly included in the “product” definition (in contrast to the old directive’s ambiguity)
- Strict liability: the injured party need not prove fault — only defect + damage + causation
- Evidence / burden-of-proof relief: for “technically complex” products such as AI, member-state courts may order disclosure of evidence, and may presume causation where disclosure cannot be obtained
The AI Liability Directive was withdrawn by the Commission in 2025-02; Directive 2024/2853 absorbs most of its functions.
Direct impact on AI
Section titled “Direct impact on AI”Who may bear liability
Section titled “Who may bear liability”- Manufacturers / producers (including developers of AI models)
- Importers / authorized representatives
- Fulfillment service providers
- Economic operators who substantially modify the product (downstream parties that make material modifications to a model)
Defect determination (Article 7)
Section titled “Defect determination (Article 7)”Takes into account:
- Intended use and reasonably foreseeable use
- Reasonable expectations of safety (including instructions and reasonably foreseeable misuse)
- Compliance with regulatory standards such as the AI Act → interacts with AI Act compliance
- Behavior after self-learning / updates (Article 7(2)(c) expressly covers machine-learning-driven changes after placement on the market)
Burden of proof relief (Articles 10, 11)
Section titled “Burden of proof relief (Articles 10, 11)”- Courts may order the defendant to disclose evidence (technical documentation, training-data metadata)
- If the defendant refuses disclosure → defect may be presumed
- Where the litigation is “excessively complex” (notably including AI cases) → defect or causation may be presumed
Limitation periods (Articles 16, 17)
Section titled “Limitation periods (Articles 16, 17)”- Procedural limitation: 3 years from when damage was or should have been known
- Absolute limitation after product placement: 10 years
- Latent injury (e.g., long-term exposure): 25 years
Relationship with the AI Act
Section titled “Relationship with the AI Act”- AI Act: ex ante compliance (before placement on the market)
- This Directive: ex post liability (after damage occurs)
- AI Act compliance does not exempt from this Directive’s liability (though it is important evidence in defense)
Withdrawal of the EU AI Liability Directive
Section titled “Withdrawal of the EU AI Liability Directive”The originally proposed AI Liability Directive (on coordination of fault-based liability) was withdrawn from the Commission’s legislative agenda in 2025-02. Official reasoning: Directive 2024/2853 already covers most substantive issues, and member states were deeply divided on a unified fault-liability mechanism. Practical effect: AI fault-based liability is left to national law, with attendant fragmentation risk.
Text and translations
Section titled “Text and translations”| Language | Source | Link |
|---|---|---|
| English (original) | EUR-Lex | eur-lex.europa.eu |
| Multilingual versions | EUR-Lex | 24 official languages |
Version history
Section titled “Version history”| Date | Event |
|---|---|
| 2022-09 | Commission proposal |
| 2024-03 | Parliament adoption |
| 2024-10-10 | Council adoption |
| 2024-11-18 | Published in Official Journal |
| 2024-12-08 | Entry into force |
| 2026-12-09 | Member-state transposition deadline |
| 2025-02 | AI Liability Directive withdrawn |