In 2021 we wrote about the European Commission’s draft regulation on artificial intelligence (AI) whose rules on AI systems include a risk-based approach. This means that any AI systems that are recognised as unacceptable risk systems are not permitted in the EU at all. For high-risk AI systems the draft regulation lays down stringent conditions that must be met before they can be distributed in the EU market. While preparations are being made for applying the AI regulation, in September 2022 the Commission presented a draft directive on non-contractual AI liability rules (the “draft liability directive”), which together with the draft AI regulation and AI product safety rules will create a framework for distributing AI systems in the EU market. This article explores what the draft liability directive requires of companies wishing to distribute AI systems on the market.

Why the liability directive was drafted

The member states’ current legislation on liability and loss calculation is mainly based on an injured party’s obligation to prove unlawful conduct and causation between such conduct and a loss. Considering AI’s nature and complexity, victims may find it difficult or overly expensive to identify the defendant and make a claim for recovery of losses resulting from AI products or services. Some of the member states were planning to draft their own national law on AI system liability matters, yet this solution might cause a fragmented approach at EU level.

To eliminate legal uncertainty and create a legal framework that promotes AI’s development and predictability, including liability matters, it was decided to draft a liability directive.

What the draft liability directive envisages

The directive will be restricted to civil liability for losses caused by AI systems, with possible criminal liability outside its scope. The directive will apply to jurisdictions where the national law requires the reimbursement of losses resulting from malicious conduct, through gross or slight negligence, so the member states should have no difficulty passing the directive’s requirements into their national law that is currently effective. The directive will not be retrospective – its rules will be restricted to claims made after it is passed into a member state’s national law.

To make it easier for the claimant to bring a claim, the member states will have to ensure their courts may order the (potential) defendant to release evidence to the (potential) claimant about a particular AI system that could have caused a loss. The courts will make this order only if the (potential) claimant is able to demonstrate that all possible attempts to obtain such evidence directly from the (potential) defendant have been made.

One of the most difficult tasks a claimant usually faces is to demonstrate causation between the defendant’s conduct and the resultant loss – this could be especially difficult with an AI system. To make it easier to prove causation, the liability directive actually introduces a presumption of the defendant’s guilt. While the draft does not define the defendant, this is likely to be the AI system’s creator who developed it for distribution on the market. The courts will be authorised to presume the defendant is guilty if the following preconditions are all met:

1. The claimant has demonstrated the defendant’s guilt in causing a loss through lack of care.
2. There are grounds to believe the defendant’s lack of care affected an AI system’s operation.
3. The claimant has demonstrated the AI system’s operation caused him the loss.

When the liability directive could take effect

It is expected to take several years for the draft to be approved and passed into the member states’ national laws. However, combined with other drafts on AI, the liability directive indicates the EU’s firm stance and clear vision on AI systems, which are to be adopted in several areas. To gear up for the new requirements properly, companies are advised to monitor the whole process of AI rules being approved and taking effect across the EU.