EU AI Liability Directive

Article 4(1) provides that, subject to the conditions in Article 4(2)-(3), national courts presume, for the purposes of applying liability rules to fault-based non-contractual civil-liability claims for damage caused by an AI system, the causal link between the defendant's fault and the output produced by the AI system, or the failure of the AI system to produce an output, where: (a) the claimant has demonstrated, or the court has presumed, the fault of the defendant or a person for whose conduct the defendant is responsible consisting in the non-compliance with a duty of care under Union or national law intended to protect against the damage that occurred; (b) it can be considered reasonably likely, based on the circumstances of the case, that the fault has influenced the AI-system output (or absence thereof); and (c) the claimant has demonstrated that the output (or absence thereof) ga

• Mapping of duties of care under Union or national law that apply to the entity's AI systems (AI Act provider/deployer obligations, sectoral obligations, GDPR Article 32)
• Defence file showing that, even if fault is established, it did not influence the AI-system output (technical chain-of-causation evidence)

• No mapping of which AI-Act obligations create 'duties of care' that the 4(1) presumption can hook into
• No chain-of-causation documentation linking fault to output for any deployed AI system

Article 4(2) provides that for claims against a provider of a high-risk AI system (or a person subject to provider obligations under the AI Act) the condition in Article 4(1)(a) is to be regarded as met only where the claimant has demonstrated that the provider or such person failed to comply with specified AI-Act obligations, in particular: risk management (Art. 9), data and data governance (Art. 10), technical documentation (Art. 11), record-keeping (Art. 12), transparency (Art. 13), human oversight (Art. 14), accuracy/robustness/cybersecurity (Art. 15) or corrective actions (Art. 21). The article also lists analogous obligations for deployers (former 'users' in the proposal text) of high-risk AI systems.

• Compliance evidence for each AI-Act obligation enumerated in Article 4(2) for the entity's high-risk AI systems
• Gap analysis showing which obligations could expose the entity to the 4(1) causal-link presumption if the proposal were enacted

• No conformance evidence to AI-Act Articles 9-15 and 21 for high-risk systems
• Treating Article 4(2) of AILD as duplicative of AI-Act conformity (it is not - it is the procedural causation bridge)

Article 4(3) provides that for claims involving non-high-risk AI systems the 4(1) causal-link presumption applies only where the national court considers it excessively difficult for the claimant to prove the causal link. The aim is to avoid extending the presumption broadly to all AI systems while still ensuring that opacity does not block recovery in genuinely intractable cases.

• Documentation that distinguishes non-high-risk AI systems from high-risk for the entity's AI inventory
• Internal note on how the 'excessively difficult' threshold has been interpreted in any test-litigation

• Assuming the 4(1) presumption automatically applies to all AI systems regardless of risk classification
• No internal classification of AI systems against the AI-Act high-risk criteria

Article 4(6) provides that for claims against a defendant who used the AI system in the course of a personal, non-professional activity the 4(1) presumption applies only where the defendant materially interfered with the conditions of operation of the AI system or where the defendant was required and able to determine the conditions of operation but failed to do so.

• Documentation distinguishing the entity's professional deployments from personal-use scenarios
• Records of operating-condition controls available to the deployer

• Treating personal-use deployments identically to professional-use for liability-presumption purposes

Article 4(7) provides that the 4(1) causal-link presumption is rebuttable. The defendant is to have the right to rebut the presumption, including by showing that a cause other than the fault, or other than the AI system's behaviour, was the cause of the damage.

• Defence playbook setting out the technical and procedural evidence the entity would marshal to rebut the 4(1) presumption
• Independent expert and forensic-engineering arrangements for chain-of-causation analysis

• No rebuttal playbook, leaving the entity exposed if the 4(1) presumption attaches

Article 3(1) provides that Member States ensure that national courts are empowered, upon a request by a potential claimant who has previously asked a provider, a person subject to provider obligations under the AI Act, or a user of a high-risk AI system to disclose relevant evidence at their disposal about that specific high-risk AI system suspected of having caused damage, and where that request has been refused, to order the disclosure of such evidence from those persons.

• Inventory of evidence that the provider/deployer of a high-risk AI system holds and could be ordered to disclose (technical documentation, logs, deployment configuration, post-market monitoring records)
• Pre-litigation request channel for potential claimants and documented refusal records
• Legal-hold and preservation procedures keyed to potential claims

• No procedure to receive and process a pre-litigation evidence request from a potential claimant
• Evidence destruction policies that would prevent disclosure of high-risk-AI documentation
• Logs and technical documentation kept only by sub-processors with no contractual disclosure path

Article 3(2) provides that for evidence requests against persons that are not defendants the claimant must present facts and evidence sufficient to support the plausibility of a claim for damages. Disclosure or preservation orders against non-defendants are subject to the same proportionality test as Article 3(4).

• Plausibility-of-claim memorandum standard for the entity when responding to disclosure requests
• Documentation of the entity's role (defendant/non-defendant) for each AI system in scope

• No internal triage between disclosure requests targeting defendants vs non-defendants
• Disclosure responses that ignore the plausibility threshold and disclose excess material

Article 3(3) provides that national courts may also order the preservation of evidence about specific high-risk AI systems suspected of having caused damage. Preservation orders may be issued independently of, or together with, a disclosure order.

• Legal-hold playbook covering AI-system technical documentation, automated logs (per AI Act Article 12), deployment configurations and post-market monitoring records
• Audit trail of preservation order receipt and execution

• No legal-hold capability for AI-system logs (which under AI Act Article 12 may rotate or be deleted)
• Preservation that omits the deployment-environment configuration of the specific incident

Article 3(4) requires that courts limit disclosure or preservation to that necessary and proportionate to sustain a claim or potential claim, taking into account the legitimate interests of all parties, including third parties, in particular the protection of trade secrets within the meaning of Directive (EU) 2016/943 and confidential information. Member States ensure that procedural remedies are available against such orders, including measures to preserve confidentiality of disclosed material.

• Trade-secret protection plan for AI-system technical documentation (filtered/redacted versions; in-camera review; confidentiality clubs)
• Documented procedural-remedy paths the entity will invoke against disproportionate disclosure

• Disclosure responses that release source code, model weights or training-data details without proportionality assessment
• No procedure to assert Directive (EU) 2016/943 trade-secret protection during disclosure

Article 3(5) introduces a procedural presumption: where a defendant fails to comply with an order to disclose or preserve evidence at its disposal, the national court is to presume the defendant's non-compliance with a relevant duty of care, in particular the duty of care that the disclosed evidence was intended to prove, that the disclosure was intended to establish for the purposes of the relevant damages claim. The presumption is rebuttable.

• Internal escalation procedure when an entity is ordered to disclose AI-system evidence (avoid 3(5) presumption attaching)
• Defence file to rebut the 3(5) presumption (good-faith refusal, partial disclosure, asserted privilege)

• No procedure to comply with disclosure orders, exposing the entity to the 3(5) presumption
• Failure to document the basis for any refusal of disclosure

The Commission withdrew the AI Liability Directive proposal in February 2025 as part of its 2025 work-programme prioritisation, citing the entry into force of the AI Act (Regulation (EU) 2024/1689) and the revised Product Liability Directive ((EU) 2024/2853) which together cover AI-related harms (AI as a product, defectiveness presumption on disclosure failure under the revised PLD, and the AI-Act conformity regime). Some Member States and consumer-protection stakeholders advocated reintroducing the proposal; as of the corpus retrieval date the Commission had not done so. This corpus node records the withdrawn proposal text for reference and traceability; corpus status is 'referenced' (not enacted EU law).

• Internal reference noting that AILD is not enacted EU law
• Watch-list entry for any successor proposal the Commission may bring

• Treating AILD as binding when assessing the entity's EU AI-liability obligations (the binding instruments are the AI Act and the revised PLD)

Article 5 of the proposal required the Commission to evaluate the application of the Directive five years after its transposition deadline and present, where appropriate, proposals for amendment. A targeted review was to consider whether further rules were needed on strict liability, mandatory insurance and the scope of damages covered, taking into account market and technological developments. (The proposal was withdrawn in February 2025, so this review obligation did not enter into force.)

• Reference register of the entity's exposure to any successor instrument the Commission may introduce in lieu of AILD

• Treating the Article 5 review as a binding obligation on the entity (it would have bound the Commission, and the proposal is withdrawn)

Article 6 of the proposal would have amended Annex I to Directive (EU) 2020/1828 (Representative Actions for the Protection of the Collective Interests of Consumers) to add the AILD to the list of instruments under which qualified entities can bring representative actions. This would have permitted collective actions for AI-related fault-based damages claims under the harmonised representative-actions regime.

• Internal note on the entity's exposure to representative actions under Directive (EU) 2020/1828 for AI-related claims (irrespective of AILD's withdrawal)

• Assuming AI-related collective redress is unavailable in the EU because AILD was withdrawn (Directive (EU) 2020/1828 continues to operate; sectoral instruments still feed it)

Article 1 of the proposal sets the subject matter and scope. The Directive applies to non-contractual fault-based civil-law claims for damages caused by AI systems where such claims are brought under national fault-liability regimes. It harmonises specific procedural rules (disclosure of evidence and the presumption of a causal link in Article 4) without prejudice to other Union rules, Member State strict-liability rules and the right of injured persons to claim under other regimes.

• Internal mapping of the proposal to the entity's product/service that could trigger non-contractual civil-liability claims under national law
• Boundary statement linking AILD scope to the AI Act and the revised Product Liability Directive ((EU) 2024/2853)

• Treating AILD as enacted EU law (it is a withdrawn proposal, not in force)
• Conflating AILD (non-contractual, fault-based) with the revised PLD (strict product liability)

Article 2 of the proposal supplies definitions, expressly aligned with the AI Act (Regulation (EU) 2024/1689). Key definitions cover 'AI system', 'high-risk AI system', 'provider', 'user' (now 'deployer' under the AI Act final text), 'claimant', 'potential claimant', 'fault' and the categories of persons entitled to bring a claim, including those who have succeeded to or have been subrogated to the rights of the injured person and those acting on behalf of one or more injured persons under Union or national law.

• Definitions glossary aligning the entity's internal terminology to the AILD/AI-Act terms
• Documentation of subrogation/representative-action standing relevant to the entity

• Using stale 'user' terminology where the AI Act final text uses 'deployer'
• Defining 'high-risk AI' inconsistently with the AI Act Annex III categories

Article 7 of the proposal required Member States to bring into force the laws, regulations and administrative provisions necessary to comply with the Directive by a date 24 months after its entry into force, and to communicate the text of those provisions to the Commission. (The proposal was withdrawn before adoption; no transposition deadline ever applied.)

• Reference note that the Article 7 transposition obligation was never triggered (proposal withdrawn February 2025)

• Tracking imaginary AILD transposition dates in compliance calendars

Article 8 of the proposal provided that the Directive would enter into force on the twentieth day following its publication in the Official Journal of the European Union. (The proposal was withdrawn before adoption; the Directive never entered into force.)

• Reference note that AILD never entered into force

• Tracking an imaginary AILD entry-into-force date

Article 9 of the proposal provided that the Directive was addressed to the Member States. (The proposal was withdrawn before adoption; no Member State was ever an addressee of a binding AILD.)

• Reference note that AILD was never addressed to Member States in binding form

• Treating Member States as having received an AILD obligation