The proposed EU AI Liability Directive makes it easier for people harmed by AI systems to bring civil claims, by easing their burden of proof. This scorer turns the directive’s key levers — risk tier, your role, the harm type, and your disclosure posture — into a single relative exposure figure so product and legal teams can compare scenarios.
How the directive changes the litigation landscape
Before the AI Liability Directive proposal, a claimant harmed by an AI system faced a practical barrier: it was very difficult to prove that a specific AI output caused a specific harm, especially when the internal workings of the system were opaque. The directive addresses this barrier in two ways.
Article 3 — Disclosure of evidence lets a national court order a provider or deployer to disclose evidence about a high-risk AI system where a claimant has a plausible case that the system caused them damage. If the defendant refuses or is unable to comply with a disclosure order, the court may presume that the defendant was non-compliant with a relevant duty of care — materially shifting the balance of the case. This creates a strong incentive to maintain complete, accessible documentation rather than relying on opacity.
Article 4 — Rebuttable presumption of causation applies when a claimant shows that a defendant was at fault and it is reasonably likely that the fault influenced the AI system’s output. In those circumstances, the court presumes a causal link between the fault and the damage — reversing the burden of proof so the defendant must demonstrate no causation rather than the claimant demonstrating it. This presumption applies most readily to high-risk AI systems as defined by the AI Act.
How it works
The directive eases the claimant’s burden through two main mechanisms, both of which this tool weights:
- Disclosure of evidence (Article 3) — courts can order disclosure of evidence about high-risk AI systems. Refusing a disclosure order triggers a rebuttable presumption of non-compliance with a relevant duty of care.
- Rebuttable presumption of causation (Article 4) — once fault is shown and it is reasonably likely the fault influenced the AI output, causation between the fault and the output is presumed, shifting the burden to the defendant.
The presumption applies more readily to high-risk AI systems as classified by the AI Act, and is qualified for non-high-risk systems. The tool scores four inputs — risk tier, deployment role, harm type, and disclosure posture — sums their weights, and normalises the total to a 0 to 100 exposure figure with a band.
Illustrative scenarios
For example: a high-risk system (EU AI Act Annex III category) operated by a professional deployer, where a user suffers a fundamental-rights harm and disclosure is contested, scores in the high band. The tool notes that the Article 4 presumption applies readily for high-risk systems and that contested disclosure invites adverse inferences.
By contrast: a limited-risk system in an informational use case, where harm is minor and documentation is complete and readily available, scores in the low band — with the disclosure component minimal because there is no reason to resist a disclosure order.
Tips and notes
The directive is a proposal whose final form may differ, so treat the score as a relative comparison between scenarios rather than an absolute liability estimate. Complete logging, documentation, and readiness to disclose are the single most effective way to reduce the disclosure-driven component of exposure. If you are building a high-risk system, invest in audit trails and documentation as a risk management measure rather than a compliance cost — the ability to disclose cleanly is valuable protection even if the directive’s final text differs from the proposal.