Directive (EU) 2024/2853 extends strict liability from defective products to software and artificial intelligence. This article explores the specific implications for those who develop, distribute, or integrate software — including AI systems — as part of their business activities. For a general overview of the directive, including definitions, obligated parties, and the transitional regime, consult the introductory guide to Directive (EU) 2024/2853.

Software is a product: what it means in practice

The directive formally equates software with a physical product for the purposes of strict liability. This includes operating systems, firmware, applications, computer programs, and artificial intelligence systems, regardless of the distribution channel: local installation, cloud, SaaS, or integration into a physical device.

Free and open-source software developed or distributed outside of a commercial activity, raw source code, and multimedia files such as e-books are excluded. For a detailed analysis of the digital products covered, see digital products and Directive (EU) 2024/2853.

The practical consequence is that the injured party does not need to prove the manufacturer’s fault: it is sufficient to demonstrate the defect, the damage, and the causal link.

Post-sales liability and security updates

The software manufacturer’s liability does not end with the distribution of the product. As long as the manufacturer maintains the ability to provide updates — directly or through third parties — the product remains under their control.

This means that omitting necessary security updates can render the product defective even if it was compliant at the time of distribution. Liability also extends to defects introduced by subsequent updates or modifications authorized by the manufacturer itself.

For organizations looking to verify their exposure, a periodic Vulnerability Assessment allows for the identification of known vulnerabilities before they become a source of liability.

Artificial intelligence: machine learning and presumption of defect

The safety assessment of an AI system takes into account its ability to learn and acquire new capabilities after being put into service. The manufacturer remains liable even if the damage results from unexpected behavior produced by machine learning.

In cases where technical or scientific complexity makes it excessively difficult for the victim to prove the defect or the causal link, the court may presume the defect or the causality if the victim demonstrates that one or the other condition is likely. This significantly lowers the threshold for accessing compensation for damages caused by opaque or complex AI systems.

Disclosure of evidence and trade secrets

In the event of litigation, courts may order software houses to disclose relevant evidence: technical documentation, system logs, design specifications. Protection measures for trade secrets are provided, but the refusal to produce the requested evidence exposes the party to the risk that the judge will automatically presume the product’s defect.

This makes document management and development lifecycle traceability a relevant element not only for software quality but also for legal defensibility.

Data damage: what is compensable

The directive recognizes the right to compensation for the destruction or corruption of data caused by a defective product. Compensation covers the material costs of recovery or restoration, but it applies only to data not used exclusively for professional purposes and only for natural persons.

Companies as legal entities and damages to goods or data used exclusively for professional purposes remain excluded from the directive’s regime: for these cases, ordinary contractual and non-contractual liability regimes continue to apply.

Impact on micro and small software enterprises

If a defect is attributable to a software component provided by a micro or small enterprise, the manufacturer of the final product may contractually waive the right of recourse against the small software house. This clause protects SMEs from unsustainable litigation between professional operators, while leaving the final consumer’s right to compensation from the product manufacturer intact.

Frequently Asked Questions

• Is a mobile app distributed via a store subject to the directive?
• Yes. Mobile applications fall under the definition of software-product regardless of the distribution channel. If the app is distributed as part of a commercial activity and causes damage due to a defect, the manufacturer can be held strictly liable.
• Is a SaaS system considered a product or a service?
• Software provided via SaaS falls within the scope of the directive as a product. Digital services necessary for the functioning of the product are treated as components of it, so the distinction between product and service does not exclude the applicability of the regime.
• Does the liability also apply to open-source software?
• Free and open-source software developed or distributed outside of a commercial activity is excluded. However, if open-source software is integrated into a commercial product or distributed as part of an economic activity, the manufacturer of the final product is liable for defects, including those deriving from open-source components.
• When does the presumption of defect for AI systems trigger?
• The presumption applies when the victim demonstrates that the defect or causal link is likely, but technical or scientific complexity makes full proof excessively difficult. In these cases, the judge may presume the defect or causality, effectively shifting the burden of proof onto the manufacturer.
• What is the risk for a company that does not provide security updates?
• As long as the manufacturer maintains the ability to update the product, the omission of necessary updates to correct known vulnerabilities can make the product defective under the directive. Damage caused by that uncorrected vulnerability can therefore give rise to strict liability.
• Can corporate clients take action under the directive?
• Legal entities (companies) are not among the subjects protected by the directive. Damages to goods or data used exclusively for professional purposes are excluded. However, companies may still take action based on ordinary contractual or non-contractual liability regimes, which remain unaffected.

Useful resources

• General guide to Directive (EU) 2024/2853 — complete overview of subjects, definitions, and the transitional regime.
• Digital products and Directive (EU) 2024/2853 — analysis of covered digital goods and exclusions.
• Vulnerability Assessment and Directive (EU) 2024/2853 — how periodic checks reduce regulatory exposure.
• Penetration testing and manufacturer liability — the role of offensive testing in managing regulatory risk.