Alerter by Lucy McCormick: New EU liability rules on AI
On 28 September 2022, the European Commission published the AI Liability Directive, aimed at adapting non-contractual civil liability rules to AI. It includes two eye-catching proposals aimed at putting claimants in AI disputes on similar footing to claimants in non-AI disputes: a ‘presumption of causality’ and a right to evidence.
SOME CONTEXT: RECENT EU DEVELOPMENTS
In a 2020 survey, concerns about liability ranked among the top three barriers to the use of AI by European companies.
It is widely accepted that national liability rules, in particular those based on fault, are generally not well-suited to handling liabilityclaims for damage caused by AI-enabled products and services. Under such rules, claimantsneed to prove a wrongful action or omission by a person who caused the damage. The specificcharacteristics of AI, including complexity, autonomy and opacity (the so-called “black box”effect), often make it difficult or prohibitively expensive for claimants to identify the liableperson and prove the requirements for a successful liability claim. In particular, claimants are likely to incur very significant up-front costs compared to cases not involving AI.
Against that background, a coordinated approach on AI law has been a priority for the European Commission. It has been seeking to support the roll-out of AI in Europe though three initiatives:
- The “AI Act”, a legislative proposal laying down horizontal rules on artificial intelligence systems. This April 2021 proposal set out rules to ensure AI systems meet high safety requirements, including logging by design and cybersecurity features. This is similar to what other EU safety legislation does for other products, such as machinery or radio equipment.
- A revision of product safety rules generally; and EU rules to address liability issues relating to AI systems.
This latter workstream has resulted in two linked policy initiatives:
- A proposal for the revision of the Product Liability Directive. This Directive covers producer’s no-fault liabilityfor defective products; and
- A proposal for a Directive on adapting non-contractual civil liability rules to AI, known as the AI Liability Directive. This covers mainly fault-based claims.
It is this last initiative, the AI Liability Directive, which is the primary subject of this Alerter.
WHAT DOES THE AI LIABILITY DIRECTIVE SAY?
The AI Liability Directive complements and modernises the EU civil liability framework, introducing for the first time rules specific to damages caused by AI systems. The Directive introduces two main measures: the so called ‘presumption of causality’, which will relieve victims from having to explain in detail how the damage was caused by a certain fault or omission; and the access to evidence from companies or suppliers, when dealing with high-risk AI. The AI Liability Directive adopts some of the concepts and terminology from the AI Act, including the distinction between “high-risk AI systems” (for example those with safety or fundamental rights implications) and other AI.
Disclosure of evidence and rebuttable presumption of non-compliance
Article 3 of the Directive provides that a court may order the disclosure ofrelevant evidence about specific high-risk AI systems that are suspected of having caused Requests for evidence are addressed to the provider of an AI system, a person who issubject to the provider’s obligations under the AI Act, ora user pursuant to the AI Act. The requests must be supported by facts and evidencesufficient to establish the plausibility of the contemplated claim for damages. Such an order may only be made if the claimant has undertaken all proportionate attempts at gathering the relevant evidence from thedefendant.
Where a defendant fails to comply with such an order, the Court shall presume the defendant’s non-compliance with a relevantduty of care (though such presumption can be rebutted, for example by proving that a different cause was at fault). This of course is a strong incentive for compliance.
Rebuttable presumption of a causal link in the case of fault
Article 4 of the Directive creates a rebuttable presumption of a causal link. Such a presumption shall come into place if the following conditions are met:
- the claimant has demonstrated (or the court has presumed pursuant to Article 3) the fault of the defendant, or of a person for whose behaviour the defendant isresponsible, consisting in the non-compliance with a duty of care laid down in Unionor national law directly intended to protect against the damage that occurred;
- it can be considered reasonably likely, based on the circumstances of the case, thatthe fault has influenced the output produced by the AI system or the failure of the AIsystem to produce an output; and
- the claimant has demonstrated that the output produced by the AI system or thefailure of the AI system to produce an output gave rise to the damage.
More detailed requirements for satisfying the first condition (i.e. fault) are set out in respect of AI systems classed as ‘high-risk’. Claimants would need to show that providers or users of such systems have failed to comply with the obligations arising under the AI Act. These include, for example, requirements to train AI on appropriate models and data sets, and requirements to allow effective human oversight of the AI in use.
The defendant is able to argue that a court should not apply the presumption of causal link if in fact the available evidence and expertise is accessible to the Claimant (with different standards for this set out for high risk and non-high risk AI).
The industry is unlikely to greet the disclosure and burden of proof provisions with open arms. However, it is clear that the AI Liability Directive (and other EU AI proposals) are intended to strike a balance between the needs of victims and of businesses. It is for this reason that – rather than a wholesale reversal of the burden of proof – the Commission has chosen the comparatively mild tool of rebuttable presumptions.
The above proposals are presently pending to be adopted by the European Parliament and the Council. Post-Brexit, it remains to be seen whether there is appetite in this jurisdiction for similar provisions.
This article is available to download as a PDF below.
To subscribe to Henderson Chambers news, alerters and updates please click here.
Download Alerter by Lucy McCormick: New EU liability rules on AI