Definition: AI Liability Directive
The AI Liability Directive was a proposed EU law intended to ease the burden of proof for people harmed by AI systems in fault-based civil claims, through a rebuttable presumption of causality and new evidence-disclosure rights, before the European Commission withdrew the proposal in October 2025.
Core characteristics of the AI Liability Directive
Unlike a strict-liability product law, the AILD targeted the evidentiary problem in existing national tort systems: proving that a specific AI system’s fault caused a specific harm.
- Covered non-contractual, fault-based claims against AI providers, deployers and users
- Introduced a rebuttable presumption of causality for high-risk AI systems
- Gave courts power to order disclosure of technical documentation before trial
- Left the underlying national tort law, such as negligence standards, untouched
AI Liability Directive vs. AI Liability and the Product Liability Directive
AI liability is the broad umbrella covering every legal route by which AI-caused harm reaches a provider or operator, including tort law, the EU AI Act, and data protection law. The Product Liability Directive is a separate, binding EU instrument imposing strict, no-fault liability on manufacturers when software or AI counts as a defective product. The AILD was narrower and complementary to both: it addressed only fault-based claims, such as professional negligence or a flawed deployment decision, that fall outside strict product liability entirely.
Importance of the AI Liability Directive in enterprise AI
Its withdrawal matters because the gap it targeted, proving fault and causation against an opaque AI system, has not been filled by another instrument. Bitkom’s September 2026 Mittelstandsstudie found that 66% of German companies name legal hurdles and uncertainty as a barrier to AI adoption, and unresolved liability questions are a direct driver of that figure.
Methods and procedures for the AI Liability Directive gap
Since the directive never entered force, companies manage the resulting gap with existing tools rather than a dedicated AILD process.
Relying on the Product Liability Directive where it applies
Where an AI system is embedded in or sold as a product, the revised Product Liability Directive now carries most of the practical weight the AILD would have added for defect-based claims, including its own disclosure rules.
Falling back on national tort law for everything else
Purely professional AI services, such as AI-assisted advice that never reaches the market as a product, remain governed by ordinary national tort rules, which still require claimants to prove fault and causation without any AI-specific presumption.
Covering the residual gap with insurance and oversight
Companies increasingly close the remaining exposure through dedicated AI liability insurance and documented human review of AI decisions, rather than waiting for a legislative fix.
Important KPIs for AI Liability Directive readiness
Because there is no AILD compliance regime to measure, readiness here means being prepared for whichever claim route actually applies.
Documentation and evidence metrics
- Technical documentation retrieval time: under 5 business days
- AI decision logs retained: at least 6 months, longer for high-risk systems
- Human oversight records: logged for 100% of consequential AI decisions
- Vendor disclosure clauses: present in 100% of AI procurement contracts
Legal exposure tracking
Legal teams should track which AI use cases could trigger Product Liability Directive strict liability versus fault-based national tort claims, since the evidentiary standard differs sharply between the two.
Policy awareness metrics
Few in-house legal teams outside large enterprises actively track EU legislative withdrawals; a short annual review of pending AI-specific civil liability proposals is a low-cost control.
Risk factors and controls for the AI Liability Directive gap
Three structural risks follow directly from the AILD’s withdrawal.
Harder claims for non-product AI harms
A company whose AI system causes harm through a service rather than a defective product faces claimants without any AI-specific presumption of causality to help them.
- Internal AI tools influencing hiring, credit or pricing decisions
- AI-assisted professional advice never sold as a standalone product
- Consulting-style AI deployments outside the Product Liability Directive’s scope
Legislative uncertainty
The Commission has not ruled out a future, narrower liability instrument, and some legal scholars and MEPs argue that equivalent protections should be folded into other frameworks. Companies should not assume the current gap is permanent.
Reputational exposure independent of legal outcome
Even where a claimant cannot ultimately prove causation, a publicized AI-harm dispute carries reputational cost regardless of the final judgment, which argues for documented human-in-the-loop review as a safeguard beyond strict legal necessity.
Practical example
A 140-employee tax advisory firm in Hamburg uses an AI tool to flag anomalies in client bookkeeping before filing. When the tool missed an anomaly that led to a client penalty, the client’s lawyer initially assumed an AI-specific presumption of fault would apply, based on older coverage of the AILD proposal. The firm’s counsel clarified that, since the AILD was withdrawn, the claim had to proceed under ordinary negligence rules, requiring the client to show the firm’s staff failed to exercise reasonable professional care in supervising the tool.
- Documented review log showing which anomalies staff checked manually
- Written AI use policy describing the tool’s role as advisory, not final
- Vendor contract clause confirming technical documentation would be available on request
- Early legal clarification of which liability track, tort or product, actually applied
Current developments and effects
The directive’s absence continues to generate legislative and academic debate rather than a settled replacement.
Commission’s next steps remain open
The Commission has said it will assess whether a different approach to AI civil liability is needed, without committing to a timeline.
- No formal replacement proposal has been tabled as of late 2026
- The revised Product Liability Directive remains the primary EU-level instrument for AI-caused harm
- National tort law continues to govern the gap for non-product AI services
Parliamentary and academic pushback
Some members of the European Parliament and legal scholars have criticized the withdrawal, arguing that fault-based claims against professional AI services still lack an adequate evidentiary framework.
Scope debate continues
Commentary continues to debate whether the Product Liability Directive’s product concept can realistically stretch far enough to cover every AI harm the AILD was meant to address, or whether a narrower, AI-specific fix will eventually return.
Conclusion
The AI Liability Directive was withdrawn, not replaced, and the evidentiary gap it targeted remains open for anything that falls outside the Product Liability Directive’s product-based scope. Mittelstand companies should not wait for a future EU fix: documented human oversight, clear AI use policies and solid vendor disclosure clauses are currently the most effective substitutes. Whether equivalent protections return in another form is unresolved, so legal teams should revisit this topic at least annually.
Frequently Asked Questions
Is the AI Liability Directive still in force?
No. The European Commission listed the proposal for withdrawal in its 2025 Work Programme in February 2025 and formally withdrew it in October 2025, as part of a regulatory simplification agenda. It never became binding EU law, so no company needs to comply with it.
What was the AI Liability Directive actually trying to fix?
It targeted the difficulty claimants face proving that a specific, often opaque AI system’s fault caused their specific harm, through a rebuttable presumption of causality and court-ordered disclosure of technical documentation.
If the AILD is gone, what liability rules apply to our AI tools now?
Three tracks apply depending on the case: national tort law, such as Section 823 BGB in Germany, for fault-based claims; the revised Product Liability Directive for strict liability where AI counts as a defective product; and EU AI Act administrative penalties for regulatory non-compliance.
Does this affect a Mittelstand company with under 200 employees?
Yes, the liability rules apply regardless of company size. A smaller company is less likely to face AILD-style disclosure fights since the directive never took effect, but it faces the same tort and product-liability exposure as larger firms when it deploys or builds AI.
Should we wait for a possible new EU proposal before updating our compliance approach?
No. Legal scholars and some MEPs continue debating whether a replacement instrument is needed, but there is no confirmed timeline. Document human oversight, update AI vendor contracts and align with existing EU AI Act work now rather than waiting on uncertain future legislation.
Is there funding available to help with AI liability preparation?
No dedicated EU or German funding programme covers legal liability preparation specifically, since it is a legal obligation rather than a digitalization investment. General digitalization funding for AI governance and documentation tooling can still offset some of the underlying compliance work.