New EU liability rules proposed for artificial intelligence

Alongside the proposed revision of the PLD (blogged here) the Commission recently published its proposal for a new “Directive on adapting non-contractual civil liability rules to artificial intelligence (AI Liability Directive)” (AILD). The Commission intends the AILD to supplement national rules on fault-based liability. As drafted the proposals would increase litigation risk for companies that design and/or deploy AI. You can provide feedback on the AILD up to 28 November 2022.

Further detail on the AILD and how this sits alongside the proposed PLD reforms are set out below, but in short proposed new rules under the AILD would:

  • Empower Courts to order a party that bears an obligation under the EU AI Act to disclose certain information about high-risk AI-systems. Non-compliance with such an order could lead to an easing of the burden of proof.
  • Alleviate the claimant’s burden of proof by introducing a rebuttable “presumption of causality” where, for example, certain obligations have not been met under national or EU laws.

What is it and how does it fit with the PLD and EU AI Act?

The proposed AILD is part of a two-pronged approach by the Commission to address the perceived challenges claimants face in obtaining compensation for damages caused by AI-systems. The two-pronged approach comprises:

  1. reforms under the proposed revision of the PLD for strict-liability claims, such as proposals to expand the definition of a product to include software, to include self-learning abilities in the non-exhaustive list of factors to be considered in an assessment of whether a product is defective, to extend liability to cover changes made to products after they have been placed on the market, and allowing the burden of proof to be eased in certain complex cases (like AI); and
  2. targeted amendments under the proposed AILD to the fault-based liability laws of member states.

The PLD established a strict liability (i.e. no fault) regime to enable claimants to seek compensation for defective products across the EU. The strict liability regime sits alongside fault-based regimes allowing claims for defective products such as tort (i.e. negligence) under the national laws of member states.

Claimants may bring a claim under national fault-based rules rather than the strict liability regime for certain reasons – such as where the type of damage being claimed is not recoverable under the strict liability regime, but is available under national fault based rules (e.g. damage to commercial property). Claimants may also bring a claim under national fault-based rules in the same action as a claim under the strict liability regime (i.e. as an alternative cause of action) where this is allowed by the procedural rules of the applicable member state.

The proposed AILD and reforms to the PLD are complementary to the proposed EU AI Act that is currently working its way through the EU legislative procedure. The proposed EU AI Act is preventative in that it aims to reduce risks and prevent damages, whereas the revised PLD and the new AILD aim to ensure that a claimant can obtain compensation if they have suffered damage caused by an AI-system.

What is being proposed?

The proposed AILD would apply to non-contractual civil law claims for damages caused by an AI-system brought under national fault-based liability regimes. Criminal liability is excluded.

The proposals include:

  • Introducing a mechanism to empower the Courts to order that a party who bears an obligation under the EU AI Act (i.e. a provider, user etc.) disclose certain information about high-risk AI-systems to support a claim. Non-compliance with such an order could lead to an easing of the burden of proof (a rebuttable presumption of non-compliance with a duty of care). Certain safeguards are proposed e.g. to limit access to what is necessary and proportionate and protections for trade secrets.  
  • Alleviating the claimant’s burden of proof by introducing a rebuttable “presumption of causality” where, a claimant can show that an obligation under national or EU laws relevant to the harm was not complied with, a causal link is “reasonably likely” and the output (or failure to produce an output) of the AI-system gave rise to the damage. For high-risk AI-systems, this would be tied to certain obligations under the proposed EU AI Act and taking into account the risk management system. A defendant could rebut this presumption by e.g. proving a different cause was responsible for the damage suffered.
  • The AILD would be added into scope of the upcoming representative action rules.
  • The Commission would be required to undertake a review within 5 years to assess whether there is a need to introduce no-fault lability rules for claims against operators of certain AI systems (where such rules don’t already exist at an EU level e.g. under the PLD) and mandatory insurance for the operation of certain AI-systems.
  • Definitions of “AI-system”, “high-risk”, “provider” etc. are aligned with the definitions under the proposed EU AI Act.  

The proposals only harmonise very targeted aspects and leave other matters to national law (such as calculation of damages, standard of proof, limitation periods etc.).

Why is this important?

The proposed AILD, alongside the proposed reforms to the PLD and upcoming new rules on representative actions in the EU, are set to significantly change the liability risks for all product manufacturers and suppliers who place products incorporating AI systems onto the EU market. The proposed AILD will bring broad rights for potential claimants, in many cases, to require companies to disclose documents.  It will also lead to increased exposure in litigation generally (especially class actions) for companies deploying AI, where obligations under the EU AI Act or other laws have not been complied with.

When will the new rules apply?

On our calculation, the new AILD could enter into force around 2024. It includes a proposed 2 year transition period and would apply to cases where the damage occurs after the end of the transition period – around 2026.

In terms of next steps, both pieces of draft legislation (revised PLD and AILD) are undergoing an 8-week feedback period which is open to industry and the public. Feedback on the AILD currently closes on 28 November 2022 (see here). Feedback received during this process is usually summarised by the Commission and presented to the European Council and European Parliament to consider as they review the proposal. The legislative proposals will then make their way through the ordinary legislative process involving the Council and Parliament each separately reviewing the proposal before meeting in trilogue negotiations with the Commission.

The proposed AILD is a Directive, meaning that member states will be required to transpose the requirements into their national rules. Under the current legislative proposal, member states would need to do this within 2 years after the new laws enter into force.

Where can I find out more?

  • A copy of the European Commission’s legislative proposal for the AILD can be found here
  • Questions and Answers document issued by the Commission here
  • Press release here


Jamie Humphreys