Our Expert in Switzerland

GOLD

Marcel Lanz

GOLD

AI Product Liability in Switzerland (2026): Who Is Liable for Harms Caused by Ai‑enabled Products?

By Global Law Experts

– posted 4 minutes ago

As of 6 August 2026, Swiss manufacturers, importers and distributors of AI‑enabled products face a rapidly shifting liability landscape. The revised EU Product Liability Directive, which EU Member States must transpose by December 2027, explicitly extends strict product liability to software, including AI components and post‑market updates. Although Switzerland is not an EU Member State, its economy is deeply integrated with the European single market, and the Federal Council has signalled that Swiss product‑safety rules will be kept in step with EU standards. This guide explains who bears liability under current Swiss law when an AI‑driven product causes harm, how the EU reforms affect Swiss exporters, and what practical compliance steps manufacturers and in‑house counsel should implement now.

Executive Summary: AI Product Liability Switzerland at a Glance

Under the Swiss Federal Act on Product Liability (PLA), producers bear strict liability for damage caused by defective products, no proof of fault is required. The PLA defines “producer” broadly: it captures manufacturers, importers of products into Switzerland, and any business that presents itself as the producer through branding or labelling. When an AI component is embedded in a physical product, a surgical robot, an autonomous vehicle, a smart home device, the manufacturer of that product is the primary liable party. Where a product is imported into Switzerland without the original producer being identifiable, the importer steps into the producer’s shoes.

Two EU legislative developments demand immediate attention from Swiss product teams. First, the revised EU Product Liability Directive (Directive (EU) 2024/2853) now treats standalone software, including AI systems, as a “product” and introduces burden‑of‑proof alleviation for claimants in technically complex cases. Second, the EU AI Act (Regulation (EU) 2024/1689) imposes conformity assessment, documentation, and post‑market surveillance obligations on providers and deployers of high‑risk AI systems. Swiss companies selling into the EU must comply with both regimes, regardless of their domestic obligations.

Immediate priorities:

Audit all AI‑enabled product lines against the PLA defect standard and the revised EU PLD scope.
Update risk assessments, technical files, and contractual indemnity clauses before the EU PLD transposition deadline of 9 December 2027.

Swiss Legal Landscape for Product and AI Liability

Switzerland takes a technology‑neutral, sectoral approach to regulating artificial intelligence. On 12 February 2025, the Federal Council confirmed this strategy, stating that existing laws, including product liability, tort, data protection and sector‑specific safety legislation, provide an adequate framework for addressing AI‑related risks, while reserving the option to introduce targeted regulation where gaps emerge. There is, accordingly, no standalone Swiss AI liability statute. Instead, ai liability Switzerland is governed by the interplay of four legal pillars: the Product Liability Act (PLA), the Product Safety Act (PrSA), the general tort provisions of the Swiss Code of Obligations (CO), and contractual warranties.

Swiss Product Liability Act, Strict Liability Basics

The PLA (SR 221. 112. 944) implements the original 1985 EU Product Liability Directive into Swiss law. It imposes strict (no‑fault) liability on the producer of a defective product for death, personal injury, and damage to items of personal use. A product is “defective” when it does not provide the safety a person is entitled to expect, taking into account all circumstances, including the product’s presentation, the use to which it could reasonably be put, and the time it was placed on the market. The injured party must prove the defect, the damage, and the causal link between the two.

Claims are subject to a three‑year relative limitation period from when the injured party becomes aware of the damage, the defect, and the producer’s identity, together with an absolute cut‑off of ten years from the date the product was placed on the market.

Who can be held liable if an AI‑driven product causes physical injury in Switzerland? Under the PLA, the manufacturer of the finished product is strictly liable. If that manufacturer cannot be identified, every supplier in the chain may be treated as the producer unless the supplier discloses the identity of the manufacturer or its own supplier within a reasonable time.

Sector‑Specific Rules and Safety Standards

Beyond the PLA, the Federal Act on Product Safety (PrSA, SR 930.11) requires that only safe products be placed on the Swiss market. Sector‑specific ordinances, covering medical devices, machinery, motor vehicles, radio equipment and toys, impose additional conformity‑assessment, labelling and post‑market surveillance duties. For AI‑enabled medical devices, for example, Swiss law is aligned with the EU’s Medical Devices Regulation via mutual recognition arrangements, meaning that software classified as a medical device must undergo notified‑body assessment and carry appropriate CE/CH markings. Non‑compliance with these safety standards can serve as powerful evidence of a defect in subsequent product‑liability litigation.

Types of Civil Claims and Who Can Be Sued

When an AI‑enabled product causes harm in Switzerland, claimants may pursue several overlapping theories of liability. Understanding these pathways is essential for manufacturers, importers and in‑house counsel seeking to map and manage their exposure to product liability AI claims.

Manufacturer Liability (Strict Product Liability)

The manufacturer of the finished product bears primary strict liability under the PLA. This is true even when the defect originates in a component, including an AI module, supplied by a third party. Where AI software is embedded in a physical product, industry observers expect Swiss courts to treat the software as an integral component, meaning the defect analysis focuses on the product as a whole rather than the AI in isolation. The question of manufacturer liability ai products therefore centres on whether the final product, including its AI behaviour, provides the safety the public is entitled to expect.

Importers and Distributors

Any business that imports an AI‑enabled product into Switzerland for commercial purposes may be treated as the producer under the PLA if the original manufacturer is based outside Switzerland and has no identifiable representative in the country. Distributors, wholesalers, retailers, and online marketplaces, face a more limited but still significant exposure: if they cannot identify the manufacturer or importer within a reasonable time, they can be treated as if they were the producer. This makes supply‑chain documentation and clear identification of the upstream producer a critical compliance task.

Developers, Data Providers and Integrators, Practical Exposure

Under current Swiss law, a pure software developer or training‑data provider that does not place the finished product on the market is unlikely to be caught by the PLA’s strict‑liability regime directly. However, they remain exposed to negligence claims under general tort law (Art. 41 CO) where a claimant can demonstrate that the developer failed to exercise due care, for instance, by using biased training data, omitting critical safety testing, or failing to warn of known limitations. Contractual indemnity claims between the product manufacturer and its AI supplier are also common and can shift significant financial risk upstream.

Operators and Owners of Autonomous Systems

Operators and owners of autonomous systems, such as fleet managers of self‑driving vehicles or hospitals deploying AI‑guided surgical robots, carry liability for autonomous systems through multiple channels. Keeper’s liability (Art. 58 SVG for motor vehicles), employer’s liability (Art. 55 CO), and general duty‑of‑care obligations all apply. Where an operator modifies an AI system’s parameters, disables safety features, or continues to use a system after receiving a safety notice, the operator’s own conduct may become the dominant cause of harm, potentially reducing the manufacturer’s share of liability.

When Is Software or AI a “Defect”? Strict Liability and Software Defects

Determining whether an AI system is “defective” under the PLA requires applying the consumer‑expectations test to software behaviour, a task for which there is no landmark Swiss Supreme Court precedent as of August 2026. The analysis considers what safety a reasonable user is entitled to expect, given the product’s presentation, labelling, intended purpose, and the state of scientific and technical knowledge when the product was placed on the market. For AI, this raises several novel questions around strict liability software defects.

First, model drift: an AI system that performs safely at launch but degrades over time as its operating environment changes may become defective after placement on the market. Under the current PLA, the producer can argue the product was not defective when placed on the market, a recognised defence. However, if the manufacturer retains control over the model through cloud‑based updates, early indications suggest that each update could be treated as a new placement on the market, restarting the defect analysis.

Second, training‑data defects: if a product’s AI module was trained on biased, incomplete, or contaminated data and this leads to harmful outputs (for example, a diagnostic device systematically failing to detect a condition in certain demographics), the product as a whole may be found defective. The injured party does not need to identify the precise line of code at fault, it is sufficient to prove that the product did not deliver the expected level of safety.

Third, opacity and explainability: where an AI’s decision‑making process is opaque, claimants may face difficulty proving the causal chain. The revised EU PLD addresses this through burden‑of‑proof alleviation, and industry observers expect Swiss courts to take note of this approach, particularly in cross‑border cases.

Scenario (What Went Wrong)	Likely Liable Party (Swiss View)	Practical Mitigation
Hardware failure causing injury unrelated to AI decisioning	Manufacturer of the physical product	Quality assurance, change control, hardware testing
Decision error by embedded AI (e.g., incorrect braking decision)	Manufacturer / integrator; possible operator liability if misuse	Safety‑by‑design, event logging, redundant controls
Harm caused by model trained on biased or poor data (unexpected outputs)	Developer / provider may be sued under negligence; strict liability possible if AI is integral product component	Data governance, documentation, explainability testing

Practical takeaway: Manufacturers should document training‑data provenance, model‑validation results, and post‑deployment monitoring procedures in the same technical file used for product‑safety compliance. This documentation serves as a defence in litigation and a compliance requirement under the revised EU PLD.

EU PLD and AI Act: Cross‑Border Implications for Swiss Sellers

Swiss companies that export AI‑enabled products to the European Union must comply with both the revised EU Product Liability Directive and the EU AI Act. Understanding the eu product liability directive switzerland nexus is now a business‑critical task for any Swiss exporter.

EU PLD Key Points

The revised Product Liability Directive (Directive (EU) 2024/2853), adopted in November 2024 and requiring transposition by EU Member States by 9 December 2027, marks a fundamental expansion of strict product liability. Software, including AI systems, is now explicitly included in the definition of “product. ” Manufacturers are liable not only for defects present at the time of initial market placement but also for defects introduced through software updates or the failure to provide necessary security patches. The directive introduces burden‑of‑proof alleviation: where a product is technically complex, a court may presume a defect or a causal link if the claimant demonstrates that it is likely the product was defective.

This is particularly significant for AI systems, where the inner workings of a model may be opaque even to the manufacturer.

The revised PLD also extends liability to online platforms and fulfilment‑service providers in certain circumstances, broadening the range of entities that Swiss exporters may need to indemnify or co‑defend alongside.

EU AI Act Intersection

The EU AI Act (Regulation (EU) 2024/1689) classifies AI systems by risk level and imposes graduated obligations. High‑risk AI systems, including those used in critical infrastructure, medical devices, and transport, must undergo conformity assessment, maintain technical documentation, implement risk‑management systems, and ensure human oversight. Understanding the ai act implications Switzerland is essential: Swiss providers placing high‑risk AI on the EU market must appoint an EU authorised representative and comply with all applicable requirements, mirroring the obligations that apply to EU‑based providers.

Practical Impact on Swiss Exporters and Timeline

The likely practical effect for Swiss exporters will be twofold. First, technical files and quality‑management systems must be updated to satisfy both Swiss domestic requirements (PrSA) and the EU’s enhanced PLD and AI Act standards. Second, supply‑chain contracts with EU‑based distributors and importers should be renegotiated to allocate the new liability risks, particularly the obligation to provide ongoing software updates and security patches. The transposition window closes on 9 December 2027, but early compliance is strongly recommended: several EU Member States are expected to adopt implementing legislation well before the deadline.

Immediate Compliance Checklist for Swiss Manufacturers and Importers

What practical steps should Swiss manufacturers take now to limit ai product liability Switzerland exposure? The following twelve‑point checklist covers the critical actions across four compliance phases. Each item should be assigned an owner and a target completion date.

Pre‑Market

Update risk assessments. Expand existing product‑risk assessments to include AI‑specific failure modes: model drift, adversarial attacks, training‑data contamination, and unintended emergent behaviour. Map each risk to a mitigation control.
Revise technical files. Include training‑data provenance records, model‑architecture documentation, validation and testing results, and the intended operating environment. Align with the EU AI Act’s Annex IV requirements where the product is destined for the EU market.
Conduct explainability and bias testing. For high‑risk AI systems, perform and document bias audits and explainability assessments. Retain test logs for the ten‑year PLA limitation period.

Documentation and Labelling

Revise product labelling and user instructions. Clearly state the AI system’s intended purpose, operational limits, required human oversight, and conditions that may affect performance. Inadequate instructions are a common basis for finding a product defective under the PLA.
Create or update the post‑market surveillance plan. Define metrics for monitoring AI performance after deployment, thresholds for initiating corrective action, and procedures for issuing safety notices.

Contracts

Negotiate upstream indemnities. Require AI component suppliers, training‑data providers, and cloud infrastructure vendors to warrant the safety and accuracy of their deliverables and to indemnify the manufacturer for third‑party claims arising from supplier defects.
Insert update and patch obligations. Contractually require AI suppliers to provide security patches and performance updates for a defined period, and ensure the manufacturer has the right (and the technical ability) to push updates to deployed products.
Review distributor agreements. Ensure downstream distributors understand their obligation under the PLA to identify the manufacturer or importer on request and to cooperate with any product recall.

Post‑Market and Insurance

Implement an incident‑reporting workflow. Establish a clear internal process for receiving, triaging, and reporting product‑safety incidents. Preserve all relevant evidence, model snapshots, inference logs, sensor data, immediately upon receiving a complaint.
Develop a cybersecurity and patch‑management policy. Define how in‑field software updates are authorised, tested, deployed and documented. Each update should be treated as a potential new market placement for limitation‑period purposes.
Review and renegotiate insurance coverage. Test existing product‑liability and professional‑indemnity policies against AI‑specific scenarios (see the insurance section below). Negotiate explicit coverage for software‑defect claims and product‑recall costs.
Establish a regulatory‑watch function. Assign a team member or external counsel to monitor developments in the EU PLD transposition, EU AI Act implementing measures, and Swiss Federal Council guidance on AI.

Practical takeaway: Items 1–3 and 6–7 should be prioritised immediately, as they directly affect the defences available to a manufacturer in litigation and the ability to seek contribution from upstream suppliers.

Litigation, Evidence and Remedies in Swiss Courts

While there are no landmark Swiss Supreme Court decisions addressing AI‑specific product liability as of August 2026, the procedural and evidentiary rules applicable to such claims are well established. The injured party bears the burden of proving the defect, the damage, and the causal link. In practice, the technical complexity of AI systems makes expert evidence, from forensic machine‑learning specialists, data scientists and domain engineers, central to both prosecution and defence of claims.

How to Defend AI Product Claims

Defendants in product liability AI cases have several avenues of defence under the PLA. The state‑of‑the‑art defence is available where the producer can demonstrate that the defect could not have been discovered given the scientific and technical knowledge at the time the product was placed on the market. The development‑risk defence is particularly relevant for rapidly evolving AI models, though its scope is narrow: it protects against unknowable risks, not against inadequate testing of known risk categories. Demonstrating compliance with recognised safety standards (ISO, IEC) and sector‑specific regulations strengthens a manufacturer’s position, though compliance alone is not an absolute defence under strict liability.

Causation remains a key battleground, defendants should challenge whether the AI behaviour (rather than operator misuse, third‑party interference, or environmental conditions) was the actual cause of the harm.

Evidence Checklist for Defendants

Manufacturers should preserve the following from the moment a potential claim is identified:

Model snapshot: the exact version of the AI model deployed at the time of the incident, including weights and configuration.
Training and validation data: provenance records, data‑cleaning logs, and bias‑audit results.
Inference logs: the specific inputs the system received and the outputs it generated at the time of the incident.
Change history: a full version‑control record of all software updates, patches, and configuration changes.
User instructions and warnings: the labelling and documentation provided to the end user at the time of sale and any subsequent updates.

Insurance and Risk Transfer

Traditional product‑liability insurance policies were designed for physical manufacturing defects, not for the emergent and evolving risks associated with AI. Swiss manufacturers should conduct a gap analysis of their existing coverage against the following framework:

Coverage Type	What It Covers	Recommended Action
Product liability (general)	Third‑party bodily injury and property damage from defective products	Confirm AI / software‑defect scenarios are not excluded; negotiate explicit inclusion
Product recall	Costs of recalling defective products (logistics, replacement, communication)	Extend to cover software‑only recalls (OTA update campaigns) and not just physical retrieval
Cyber / technology E&O	Losses from software errors, data breaches, system failures	Coordinate with product‑liability policy to avoid gaps and double‑deductibles for AI incidents
Directors’ and officers’ (D&O)	Personal liability of executives for inadequate risk management	Review whether AI governance failures are covered; update board reporting on AI risk

Early indications from the Swiss insurance market suggest that insurers are beginning to offer AI‑specific endorsements, but coverage terms vary widely. Manufacturers should request specimen wordings, compare sub‑limits and exclusions, and negotiate bespoke extensions where standard policies leave material gaps.

Practical Contract Clauses to Shift and Limit Liability

Contractual risk allocation is one of the most effective tools available to Swiss manufacturers managing ai product liability Switzerland exposure. While contractual clauses cannot override the PLA’s strict liability toward injured third parties, they can determine how losses are shared between parties in the supply chain.

Key clause categories to include in supplier, developer, integrator and distributor agreements:

Performance warranties: the AI supplier warrants that the system meets defined accuracy, safety and bias thresholds, measured by agreed testing protocols.
Indemnities: the supplier indemnifies the manufacturer against third‑party claims arising from defects in the AI component, subject to agreed caps and carve‑outs.
Update and maintenance obligations: the supplier commits to providing security patches and functional updates for a minimum period (ideally matching or exceeding the PLA’s ten‑year absolute limitation period).
Audit and documentation rights: the manufacturer retains the right to audit the supplier’s training data, testing records, and quality‑management system.
Liability caps and insurance floors: caps should reflect realistic claim values; require the supplier to maintain minimum product‑liability and professional‑indemnity insurance.

Practical takeaway: Standard supply‑chain contracts drafted before the AI era rarely address software updates, data provenance, or model‑drift risk. A targeted review and renegotiation of these agreements is one of the highest‑impact steps a Swiss manufacturer can take in 2026.

Next Steps

AI product liability Switzerland is a rapidly evolving area where proactive compliance delivers a measurable reduction in litigation risk and regulatory exposure. Manufacturers, importers and in‑house counsel should begin the twelve‑point compliance programme outlined above without delay, particularly items relating to risk‑assessment updates, technical‑file revisions and supplier‑contract renegotiation. For tailored guidance on Swiss product liability for AI‑enabled products, readers are encouraged to consult a qualified Swiss liability specialist through Global Law Experts.

Need Legal Advice?

This article was produced by Global Law Experts. For specialist advice on this topic, contact Marcel Lanz at Schärer Rechtsanwalte, a member of the Global Law Experts network.

Sources

FAQs

Who can be held liable if an AI‑driven product causes physical injury in Switzerland?

Under the Swiss Product Liability Act (PLA), the manufacturer of the finished product bears strict (no‑fault) liability for damage caused by a defective product. If the manufacturer cannot be identified, the importer or, ultimately, any supplier in the distribution chain may be treated as the producer. Operators may also face tort‑law claims if their conduct contributed to the harm.

Will developers or data providers face strict product liability for AI defects?

Under current Swiss law, a developer or data provider that does not place the finished product on the market is generally not caught by the PLA’s strict‑liability regime. However, they may face negligence claims under Art. 41 of the Swiss Code of Obligations if they failed to exercise due care, for example, by providing deficient training data. The revised EU PLD may also expose them if they are treated as a “manufacturer” in the EU context.

How does the EU Product Liability Directive affect AI products sold in Switzerland?

The revised EU PLD (Directive (EU) 2024/2853) treats software, including AI, as a “product” subject to strict liability and introduces burden‑of‑proof alleviation for technically complex cases. Swiss companies exporting AI‑enabled products to the EU must comply. The transposition deadline for EU Member States is 9 December 2027, but Swiss exporters should prepare now.

What immediate steps should Swiss manufacturers take to reduce liability?

Three top priorities: (1) update product‑risk assessments to include AI‑specific failure modes such as model drift and training‑data bias; (2) revise technical files to document AI architecture, training‑data provenance, and validation results; and (3) renegotiate supplier contracts to include indemnities, update obligations, and audit rights covering AI components.

Do Swiss courts have AI‑specific product liability precedents?

As of August 2026, there are no landmark Swiss Supreme Court decisions directly addressing AI product liability. The existing PLA framework, modelled on the original EU Product Liability Directive, applies on a technology‑neutral basis, and Swiss courts are expected to interpret the “defect” and “producer” concepts in light of evolving EU jurisprudence and the revised PLD.