Switzerland

The Federal Constitution contains a broad set of guarantees that may become relevant in AI matters. The most important are human dignity (Article 7), equality and non-discrimination (Article 8), protection against arbitrariness and protection of good faith (Article 9), personal liberty (Article 10), privacy and informational self-determination (Article 13), freedom of expression and information (Article 16), media freedom (Article 17), freedom of science (Article 20), freedom of the arts (Article 21), economic freedom (Article 27), and procedural guarantees, including the right to be heard and access to justice (Articles 29 and 29a). Depending on the use case, further guarantees may also be engaged, including political rights (Article 34) and sector-specific constitutional mandates.

These guarantees shape statutory interpretation, administrative discretion and judicial review. This is especially important in the public sector, where state use of AI must satisfy legality, public interest and proportionality.

Switzerland is bound by the European Convention on Human Rights, and Strasbourg case law influences Swiss debates on procedural fairness, surveillance, discrimination, freedom of expression, and algorithmic decision-making.

The leading published Swiss AI decision to date is the Federal Administrative Court’s DABUS judgment (B-2532/2024, 26 June 2025), discussed in Section 2, below. Beyond that, Switzerland does not to have a substantial body of published court decisions dealing specifically with AI.

The Federal Act on Patents for Inventions (PatA) requires that a patentable invention must be the result of human intellectual effort. AI-generated outcomes, lacking direct human contribution to the inventive process, do not fulfil this requirement. Consequently, purely AI-generated inventions cannot be patented under current Swiss law.

The most important recent authority is the Federal Administrative Court’s DABUS judgment. The court held that an AI system cannot be entered in the patent register as the inventor and that an application may not proceed without an inventor designation. At the same time, the court adopted a functionally relevant understanding of inventorship: a natural person may qualify as inventor where that person makes relevant contributions within the AI data-processing workflow, recognises the patentable result and applies for protection. Where AI is used as a tool within a human-led inventive process, patent protection remains possible, subject to the ordinary patentability requirements.

Swiss copyright law protects only works that are intellectual creations with individual character and are attributable to a human author. A purely machine-generated output with no human authorship is therefore unlikely to qualify for copyright protection under current Swiss copyright law.

Where a human uses AI merely as a sophisticated tool, the answer depends on the facts. If the human contribution remains sufficiently creative and shapes the protected expression, copyright may subsist in the output. If the allegedly creative contribution is in substance generated by the model and the human role is confined to functional prompting, selection or minor post-editing, copyright protection is far less certain.

Switzerland has no AI-specific text-and-data-mining exception comparable to the broader EU discussion and has not yet reached a consensus on training data in the context of copyright law.

Swiss trade secrets are protected through a mix of unfair competition law, contract law, employment law and, in some constellations, criminal law. Article 6 of the Federal Act against Unfair Competition prohibits the exploitation or disclosure of manufacturing or trade secrets that one has improperly obtained or otherwise learned of improperly. In practice, however, contractual confidentiality architecture remains critical.

In the AI context, the core risks are operational rather than conceptual: confidential prompts and datasets fed into third-party tools; leakage through model fine-tuning or retention settings; access rights that are too broad; insufficient segregation in enterprise environments; and inadequate restrictions on vendor use of customer inputs and outputs. For most businesses, the real protection work is therefore done through governance, vendor contracting and internal controls, not through abstract references to trade-secret doctrine alone.

The key Swiss authority is the Federal Administrative Court’s judgment B-2532/2024 of 26 June 2025 (DABUS) (see Section 2.1, above).

The Federal Act on Data Protection (FADP), in force in its revised form since 1 September 2023, is technology neutral and applies fully to AI-related processing of personal data. The essential principles therefore govern AI as they govern any other data processing: lawfulness, good faith, proportionality, purpose limitation, data accuracy, transparency, data security and accountability.

Specifically, Article 21 of the FADP governs automated individual decisions. The provision applies where a decision is based solely on automated processing and produces legal effects concerning the data subject or significantly affects the data subject. Where Article 21 applies, the controller must inform the data subject and, on request, provide an opportunity to state a point of view and to request review by a natural person, unless a statutory exception applies. The statutory trigger is narrow and should not be overstated. Not every automated process, recommendation, scoring mechanism or AI-supported workflow is an “automated individual decision” within the meaning of the FADP. The practical limitations of Article 21 of the FADP have drawn criticism. However, the Swiss Parliament confirmed that Article 21 of the FADP shall only apply in the event that the decision is based exclusively on automated processing and will not be amended for the time being.

The Federal Data Protection and Information Commissioner (FDPIC) has repeatedly stressed that — of course — current Swiss data protection law already applies to AI. In its 2025 communication, the FDPIC also stated that providers and deployers of AI systems must make the purpose, functionality and data sources of AI-based processing transparent, and that users of directly communicating language models must know whether they are interacting with a machine and whether entered data is used to improve self-learning programmes. Those statements are regulatory guidance beyond the black-letter text of the FADP, not enacted AI-specific legislation.

Given that Switzerland is not a member of the EU, Regulation (EU) 2016/679 (General Data Protection Regulation (GDPR)) does not apply directly in Switzerland (see Section 3.1, above). 

The Federal Act on the Use of Electronic Means for the Fulfilment of Official Tasks (EMBAG), in force since 1 January 2024, strengthens the Swiss open-government-data framework at federal level. As a matter of principle, federal administrative units subject to EMBAG must publish data they obtain or generate in fulfilment of their statutory tasks where the statutory conditions are met, free of charge, promptly, in machine-readable form and in an open format. The principle is commonly described as “open by default”.

The EMBAG can increase the availability of certain public sector datasets and may therefore be relevant as one lawful source of training or testing data. However, EMBAG is not a general free-data charter as its scope is limited, and statutory exceptions remain important.

Biometric data that uniquely identifies a natural person qualifies as sensitive personal data under the FADP. Its processing is therefore subject to heightened legal and organisational scrutiny. Switzerland does not currently have a dedicated horizontal AI statute that, like the EU AI Act, lists prohibited AI practices or expressly categorises biometric systems across all sectors. That does not mean biometric AI is legally unregulated. Depending on the context, the FADP, employment law, fundamental rights, cantonal police laws, employment law, administrative law legality requirements and sector-specific health or financial rules may all become relevant.

Swiss anti-discrimination law remains comparatively fragmented. Article 8 of the Federal Constitution establishes equality and non-discrimination as a constitutional guarantee, but private law implementation is selective rather than comprehensive. Statutory regimes include, in particular, the Gender Equality Act, the Disability Discrimination Act and sector-specific employment, tenancy and public law protections. Swiss criminal law also prohibits certain discriminatory conduct, for example, under Article 261 bis of the Swiss Criminal Code.

For AI, that means two things. First, discriminatory outcomes can still be legally relevant even in the absence of a dedicated AI discrimination statute. Second, Swiss law does not yet provide the kind of broad, harmonised anti-discrimination toolbox that some stakeholders would like to see for algorithmic systems. However, the Swiss Government has, to date, expressed no intentions of expanding anti-discrimination principles in view of the adaptation of the Council of Europe’s Framework Convention on Artificial Intelligence.

Switzerland does not have a comprehensive cybersecurity law. Cybersecurity is instead governed through a combination of general information security law, sector-specific regulation, contractual controls and technical standards.

The Federal Act on Information Security (ISA), effective since January 2024, mandates that federal agencies, private institutions that fulfil public tasks, and, to a certain extent, critical infrastructure providers implement robust information security practices. An amendment to the ISA, which came into force in April 2025, introduced a mandatory obligation for critical infrastructure operators to report cyberattacks to the National Cybersecurity Centre (NCSC) within 24 hours of detection. Further, there are sector-specific statutes and regulations (in particular, in the finance and healthcare sectors) that contain cybersecurity related provisions. Like the ISA, such sector-specific regulations contain generic, largely technology-neutral obligations that do not address AI specifically.

Simulations and empirical evidence show that algorithms can learn collusive strategies, leading to prices above competition, and pricing algorithms can lead to higher prices, algorithmic price discrimination and potential tacit collusion (“algorithmic collusion”). AI can intensify familiar competition law concerns such as exclusionary access practices, self-preferencing, data-based entry barriers and lock-in around critical infrastructure, computer or proprietary datasets. From a legal perspective, the biggest challenges today are that collusion without explicit agreements is not prohibited and that deep learning algorithms often work as a black box leaving no traceability of the decision-making processes. As in other jurisdictions, the analytical difficulty lies less in identifying the concern than in proving the infringement where the relevant mechanism is opaque, adaptive or decentralised.

Swiss competition law remains technology neutral. That is still true after the partial revision of the Cartel Act passed on 19 December 2025. The reform modernises merger control and amends other areas of competition law, but it does not introduce AI-specific rules on algorithmic collusion, AI-driven discrimination or AI gatekeeper power as such.

Swiss law does not recognise AI systems or AI agents as legal persons. They cannot bear rights or obligations in their own name, contract on their own behalf or own assets. Any legal effects produced through AI-assisted conduct are attributed through ordinary rules to the relevant human or legal entity.

There is currently no Swiss sector in which agentic AI enjoys independent legal capacity. Sector-specific rules may constrain or condition the use of autonomous systems, for example in finance, healthcare, aviation, road traffic or defence. But those rules regulate the conduct of companies, providers, deployers and public authorities. They do not confer legal personhood on the system.

Swiss law does not yet have an AI-specific liability regime. Liability must therefore be analysed through existing frameworks: the Product Liability Act (PLA), product safety law, contract law, tort law and sector-specific rules.

The PLA applies to defective products and imposes strict liability on the producer for death, personal injury and certain property damage. The difficult question in AI matters is not whether Swiss law has a liability regime; it is whether the relevant AI constellation falls within product liability at all. For AI embedded in tangible products, the answer is comparatively straightforward. For standalone software, cloud-based AI and service-based models, the position is less clear and may instead need to be analysed under contract and general tort law.

Strict liability under the PLA can apply where the statutory requirements are met and the relevant AI functionality forms part of a defective product. Outside that setting, liability will often shift to contract and fault-based tort analysis, unless a sector-specific strict regime applies.

Providers and deployers of AI systems embedded in physical devices are legally required to act promptly when defects or risks are discovered. This may include updates, patches, or issuing warnings and recalls. Under the PLA, liability arises, regardless of fault, if a defective AI system causes harm, making timely corrective measures critical.

Sector-specific regulations may impose additional obligations. The Ordinance on Medical Devices (MedDO) requires post-market surveillance, risk management, and field safety corrective actions for medical devices. In transportation, deployers and providers must ensure that autonomous vehicles or drones are updated, corrected, or temporarily withdrawn if safety is compromised. All corrective actions must be documented and communicated transparently to regulatory authorities and, where appropriate, to end users.

These measures collectively aim to ensure safety, compliance, and accountability, including with respect to AI-related risks.

As of early 2026, there are no reported court cases specifically addressing AI product liability in Switzerland.

The key policy development is the Federal Council’s decision of 12 February 2025 to ratify the Council of Europe AI Convention and to prepare the legal amendments required for ratification (see Section 1, above).

Switzerland will not adopt the EU AI Act, whether directly, by dynamic incorporation or by autonomous wholesale transposition. This means that Switzerland is deliberately not following the EU’s product-safety logic under the EU AI Act, under which providers of high-risk AI systems are subject to ex ante obligations before placing such systems on the market or putting them into service. Instead, Swiss law will continue, at least for the time being, to rely predominantly on ex post control: general private law, data protection law, sector-specific supervision and enforcement after risks have materialised or infringements have already occurred.

That choice has consequences. The EU model is designed to intervene upstream. It imposes compliance, documentation, risk-management, transparency and, in some cases, conformity-assessment obligations before particularly risky AI systems are deployed. The Swiss model is lighter for providers and more hospitable to innovation, but it is also less preventive. It offers fewer front-loaded legal safeguards against unsafe or rights-infringing AI and places a greater burden on affected individuals, courts and regulators to react once harm, discrimination, opacity or other adverse effects have already occurred.

In market terms, however, the EU AI Act remains highly relevant to Swiss companies. Providers established in Switzerland may fall within its scope where they place AI systems or General-Purpose AI (GPAI) models on the EU market, put them into service in the EU, or where the output is used in the EU. This is not a point of Swiss domestic law, but it is a major practical point for Swiss industry.

AI has been part of the Federal Council’s digital policy agenda for several years, but the federal approach has recently become more concrete, coordinated and implementation driven. AI is now a formal focus theme of the Digital Switzerland Strategy, covering both the regulation of AI in Switzerland and the deployment of AI systems within the Federal Administration.

The current strategy architecture includes, in particular, the Strategy Use of AI Systems in the Federal Administration, the related work on the further development of internal federal AI coordination, the legislative package to implement the Council of Europe Framework Convention on Artificial Intelligence, and a separate implementation plan for non-binding measures under that Convention. In addition, AI use is being advanced in specific administrative contexts, including office automation. The federal strategy on AI is therefore no longer limited to general principles.