Glossary · EU AI Regulation AI Liability Directive (EU AI Liability Directive (proposed, COM/2022/496))
Proposed EU directive that would harmonise non-contractual civil liability rules for damage caused by AI systems. Designed to complement the AI Act by providing victims with effective remedies. Status as of 2026: politically contested, with significant amendments under debate. Closely related to the revised Product Liability Directive (in force 2024).
## What the AI Liability Directive actually is
The AI Liability Directive (AILD, proposal COM/2022/496) is a proposed EU directive that would establish harmonised rules for non-contractual civil liability for damages caused by AI systems. It was proposed by the European Commission in September 2022 alongside the revised Product Liability Directive, as the third pillar of the EU's AI regulatory architecture:
1. **AI Act (Regulation 2024/1689)**: ex ante rules on AI system development and deployment
2. **Revised Product Liability Directive (Directive 2024/2853)**: strict liability for defective AI-enabled products
3. **AI Liability Directive (proposed)**: fault-based liability harmonisation for AI-caused damage
The AILD is the missing piece. As of 2026, its adoption is uncertain — the proposal has been significantly debated, and the Commission has signalled possible withdrawal and re-proposal in different form.
## What the AILD would do
The proposal makes two main substantive changes to existing Member State civil liability law:
### Disclosure of information
Where a claimant alleges damage from a high-risk AI system, courts could order the AI operator to disclose relevant evidence (training data, logs, system documentation). This addresses the practical problem of opaque AI systems where claimants cannot prove fault because they cannot access the relevant evidence.
The threshold is "plausible damage from AI" combined with refusal by the operator to provide the information voluntarily. Disclosure must be proportionate.
### Presumption of causality
Where a claimant has shown:
- A duty of care (under EU or national law)
- A non-compliance with that duty
- A causal link to the AI system output
The directive would create a rebuttable presumption that the non-compliance caused the damage. The AI operator would then bear the burden of rebutting the presumption.
This addresses the practical problem of proving causation in complex AI systems where the technical chain from input to harmful output is difficult or impossible to reconstruct.
## Why this is politically contested
The AILD has been the most politically debated piece of the EU AI regulatory package, for several reasons.
### Concerns about innovation impact
European AI industry associations have argued that the presumption of causality could create disproportionate liability exposure, particularly for SMEs and open-source AI developers. The argument: predictable liability is necessary for AI investment, and EU-specific presumptions could push AI activity to other jurisdictions.
### Concerns about victim protection
Civil society organisations have argued that the AI Act alone provides insufficient remedies for individuals harmed by AI systems. Without the AILD, victims face evidentiary barriers that effectively block legitimate claims.
### Interaction with Product Liability Directive
The revised Product Liability Directive (2024/2853) was adopted in October 2024 and applies to AI-enabled products. Some argue this covers significant ground that the AILD would otherwise address — making the directive redundant. Others argue the PLD covers product-level damage but not service-level or system-level damage that the AILD addresses.
### Commission re-evaluation
In early 2025 the European Commission signalled it might withdraw and re-propose the AILD in different form, given the political controversy. The status as of mid-2026 remains in flux.
## What's actually in force today
While the AILD's future is uncertain, several pieces of the broader framework are in force:
### Revised Product Liability Directive (Directive 2024/2853)
In force since October 2024. Establishes strict liability for defective products including AI-enabled products. Member States must transpose by December 2026. This significantly expands AI-product liability without requiring the AILD.
### EU AI Act (Regulation 2024/1689)
In force since August 2024 with phased applicability. While the AI Act does not directly create civil-liability remedies for individuals, it establishes baselines that strengthen the position of victims under existing national tort law.
### National AI liability law
Most Member States have their own evolving national jurisprudence on AI-caused damage under general tort principles. France, Germany, and the Netherlands have particularly active case-law development.
## What it means in practice
### For AI developers and operators
Even without the AILD, exposure to civil liability for AI-caused damage is real and increasing. Recommended preparations:
- **Maintain comprehensive logs** of AI system inputs, outputs, and decisions
- **Document training data provenance** and quality assurance processes
- **Document compliance** with relevant duty-of-care obligations (AI Act, sectoral regulations, GDPR)
- **Insurance review** — verify civil liability coverage covers AI-system-related claims
### For European businesses deploying AI
The AILD's evidentiary disclosure provisions would, if adopted, create new operational obligations to preserve and produce AI-related evidence under court order. Even absent the AILD, courts in some Member States are increasingly amenable to similar orders under existing procedural law.
### For US-headquartered AI vendors serving EU customers
EU civil liability claims against US vendors raise complex jurisdictional questions. EU procedural rules generally allow claims against US-headquartered vendors where the AI system caused harm in the EU. Coverage and insurance assumptions developed for US litigation may not translate.
### For individuals potentially harmed by AI
The current legal landscape is uneven. National tort principles apply, with significant variation across Member States. Victims with strong cases and resources can pursue claims under existing law; the AILD would standardise and lower barriers.
## AILD vs AI Act vs PLD
| Aspect | AI Liability Directive | EU AI Act | Product Liability Directive |
|--------|------------------------|-----------|------------------------------|
| Nature | Civil liability rules | Ex ante regulation | Civil liability (strict) |
| Status | Proposed; contested | In force 2024 | In force October 2024 |
| Scope | Fault-based AI damage | High-risk AI development/use | Defective AI products |
| Remedy mechanism | Court claim | Regulatory enforcement | Court claim |
| Disclosure obligations | Yes (if adopted) | Limited | Limited |
| Causation presumption | Yes (if adopted) | N/A | Yes (under conditions) |
The three instruments together (or two, if the AILD does not advance) form the EU's AI accountability architecture.
## Practical implications
- **For AI operators**: prepare for evidentiary obligations even absent AILD adoption; document everything
- **For AI vendors**: insurance and liability terms need review against current PLD and AI Act exposure
- **For deployers in regulated sectors**: existing sectoral liability regimes (medical, financial, transport) already create significant AI-deployment exposure
- **For US vendors**: EU jurisdictional reach over AI damage claims is real even absent specific harmonisation
- **For policy watchers**: AILD status remains the principal open question in EU AI regulation
The AI Liability Directive's specific provisions may or may not be adopted. But the underlying problem — civil liability for AI-caused damage in the EU — is increasingly being addressed through related instruments and case law regardless. Treating it as an active legal risk is the prudent default.
Was this helpful?
Thanks for your feedback!