On 11 February 2025, the European Commission formally withdrew its proposal for an AI Liability Directive (COM/2022/496), first presented in September 2022 as the civil-liability counterpart to the EU AI Act. The proposal was never adopted, so none of its provisions ever took effect. The withdrawal, however, does not leave a liability vacuum: liability for harm caused by artificial intelligence systems is now governed, at EU level, by the revised Product Liability Directive – Directive (EU) 2024/2853, adopted on 23 October 2024. This piece examines why the AI Liability Directive was withdrawn, what the new liability regime actually provides, and what it means in practice for businesses placing AI products or services on the EU market.

Why the AI Liability Directive was withdrawn?

The European Commission justified the withdrawal by the absence of any realistic prospect of agreement on a final text within a reasonable timeframe – the decision was taken as part of the Commission’s 2025 Work Programme. The proposal had been built around a fault-based liability model, requiring an injured party to prove a breach on the part of the AI provider or manufacturer. The Commission concluded that the revised Product Liability Directive, which explicitly brings software – including AI systems – within the definition of a product, adequately covers the same ground, rendering a separate directive unnecessary.

The new regime: the Product Liability Directive

Directive (EU) 2024/2853, adopted 23 October 2024, entered into force on 8 December 2024, but applies only to products placed on the EU market from 9 December 2026 onward. Unlike the withdrawn proposal, it establishes strict liability: an injured party need not prove fault, only that the product was defective, that damage occurred, and that a causal link exists between the two.

Key features specific to AI:

Software as a product. For the first time in the history of EU product liability law, software – including AI systems – is explicitly treated as a product rather than a service. AI providers accordingly take on manufacturer status and fall under the same liability regime as manufacturers of physical goods.

Presumption of defect. If a manufacturer refuses a court-ordered request to disclose technical documentation about an AI system, the court may presume the product was defective. This materially eases an injured party’s position in disputes involving opaque or complex AI systems, whose operation a claimant often cannot independently reconstruct or prove.

No contractual exclusion. Article 15 of the Directive expressly prohibits excluding or limiting liability for damage caused by a defective product by contract. This matters in particular for businesses operating on a software-as-a-service model, which routinely include liability-limitation clauses in their terms – such clauses will have no legal effect against liability under this directive.

Manufacturer status on substantial modification. Article 25 of the AI Act provides that a party who substantially modifies a high-risk AI system belonging to someone else assumes manufacturer status – and, consequently, liability under the Product Liability Directive – for that modification. This is particularly relevant for companies that customize or integrate third-party AI systems into their own products or services.

What stays outside the directive’s scope. Pure economic loss – the most common outcome in B2B disputes involving AI – is not compensable damage under this directive, nor are privacy infringements or discrimination. These continue to be governed by the national tort law of individual member states, which varies significantly from one jurisdiction to another.

Why this matters for Serbian businesses?

The directive is not limited to companies established in the European Union. A Serbian company whose AI product or software reaches the EU market – whether through direct sale, licensing, or EU-based users – becomes a “manufacturer” under this directive the moment the product is placed on that market, regardless of where the company or its servers are physically located. Standard liability-limitation clauses common in software-service agreements will not exclude this exposure. Technical documentation required under Annex IV of the AI Act – which businesses are already obligated to prepare for AI Act compliance – simultaneously becomes a practical defence against the presumption of defect under the Product Liability Directive.

This text is informational in nature and does not constitute legal advice.