Switzerland AI Regulation & Governance

AI Framework Law

Switzerland has no horizontal, EU-style AI statute. On 12 February 2025 the Federal Council decided against adopting a horizontal AI law, opting instead for a sector-specific, technology-neutral regulatory approach combined with ratification of the Council of Europe Framework Convention on Artificial Intelligence, Human Rights, Democracy and the Rule of Law. Switzerland signed the Convention on 27 March 2025 but has not ratified it; a public consultation on implementing legislation is expected only at the end of 2026, and a federal AI bill implementing the Convention is projected to reach public consultation in the first half of 2027.

Within the existing legal framework, the Federal Council's amendment to the Goods Control Ordinance — reportedly adopted 2 April 2025 — took effect on 1 May 2025: dual-use goods, technology and software covering artificial intelligence and other new-technology fields listed in the amended Annex 2 are subject to an export-licence requirement administered by the State Secretariat for Economic Affairs (SECO).

The non-binding Guidelines on Artificial Intelligence for the Confederation, dated 25 November 2020, continue to set the Confederation's general ethical orientation for AI use pending the sector-specific and Convention-implementing legislation described above.

Prohibited Practices

Switzerland's standing framework touching prohibited or high-risk algorithmic practices rests on the non-binding Guidelines on Artificial Intelligence for the Confederation (25 November 2020), which set ethical orientation rather than binding prohibitions on defined categories of AI use. No Swiss statute specifically prohibits categories of AI practice in the manner of the EU AI Act's prohibited-practices regime.

A legal opinion prepared for the Commission against Racism and the Commission for Women's Issues found that Swiss legal safeguards against AI-driven discrimination are inadequate, noting no Swiss equivalent to the bias-mitigation and transparency duties the EU AI Act imposes on high-risk systems. This assessment rests on a single tier-three source and carries uncertain confidence; it should be read alongside the confirmed, primary-source legal opinion on the same subject addressed in the Sectoral & Public-Sector AI module.

High-Risk Systems

Binding law: Article 21 of the Federal Act on Data Protection (FADP), in force since 1 September 2023, requires a controller that makes a decision based exclusively on automated processing — including profiling — with legal or significant effect on the data subject to inform that data subject, who may then request human review; on request, the controller must explain the rationale and criteria of the decision without disclosing trade secrets.

Regulatory expectations: the Swiss Financial Market Supervisory Authority (FINMA) published non-binding Guidance 08/2024 on 18 December 2024 setting out governance and risk-management expectations for AI use by supervised financial institutions, including maintenance of an AI inventory, risk classification, and independent review for material AI applications; the guidance applies existing principle-based supervisory rules rather than creating new AI-specific statutory duties.

Non-binding framework: the Guidelines on Artificial Intelligence for the Confederation (25 November 2020) continue to supply general ethical orientation applicable to high-risk use cases, absent a dedicated high-risk-system statute.

Transparency & Synthetic Content

No statute currently imposes AI-specific transparency or labelling duties for synthetic or deepfake content in Switzerland. The National Council rejected, on 7 May 2025, a motion that would have introduced statutory transparency or labelling rules specific to AI-generated deepfakes, concluding that existing civil and criminal law already addresses the underlying harms and relying on that existing law rather than legislating a new AI-specific regime.

A further motion, 26.3458 ('Grok und sexualisierte Deepfakes. Jetzt handeln!'), was unanimously adopted by the National Council on 16 June 2026, instructing the Federal Council to draft legislation that would require providers of generative AI to assess and mitigate the risk of sexualised deepfakes before market launch. The motion awaits consideration by the Council of States and is not yet in force; no enacted text or commencement date exists at this stage.

AI & Intellectual Property

Binding law: Article 24d of the Swiss Federal Copyright Act provides a text-and-data-mining exception covering AI training for scientific research purposes. The exception carries no opt-out mechanism for rightsholders and does not clearly extend to commercial AI training, leaving the copyright status of commercial AI-training data mining unresolved under the current statutory text.

Safety & Incident Reporting

The only recorded standing instrument addressing AI-related operational risk management for a defined Swiss sector is FINMA Guidance 08/2024, published 18 December 2024, which sets non-binding expectations for FINMA-supervised financial institutions using AI, including maintenance of an AI inventory, risk classification, and independent review for material AI applications.

Standards & Conformity

Switzerland has no dedicated AI conformity-assessment or standards-certification regime recorded. The sole relevant supervisory instrument is FINMA Guidance 08/2024, published 18 December 2024, which sets non-binding governance and risk-management expectations for FINMA-supervised financial institutions using AI, applying existing principle-based supervisory rules rather than a new AI-specific standard.

Authorities & Enforcement

Binding law: wilful breach of the Federal Act on Data Protection's information duties — including the Article 21 duty to inform data subjects of fully automated decisions — is a criminal offence, carrying a fine of up to CHF 250,000 on the responsible natural person. Separately, the Federal Council's amendment to the Goods Control Ordinance, reportedly adopted 2 April 2025 and in force from 1 May 2025, brought dual-use AI and other new-technology goods, technology and software within an export-licensing regime.

Enforcement record: the Federal Data Protection and Information Commissioner concluded its investigation under the Federal Act on Data Protection into X Corp's use of data to train Grok with a finding of compliance via X's opt-out mechanism. As at the 2026 Chambers guide, no publicly known FINMA enforcement action concerning AI had been taken. More broadly, no AI-centred enforcement action has been reported in Switzerland; that observation rests on limited underlying evidence and should be read as provisional rather than a confirmed clean enforcement record.

Sandboxes & Transitional Relief

Binding law: the only statutory sandbox mechanism in the Swiss financial sector is the technology-neutral deposit-taking exemption under Article 6 of the Banking Ordinance (BankV), in force since August 2017 and capped at CHF 1 million; it is not tailored to AI, and Switzerland has no AI-specific regulatory sandbox at the national or federal level.

Administrative testing framework: the Canton of Zürich operates an Innovation Sandbox for Artificial Intelligence, a testing environment in which businesses, public authorities and research organisations implement AI projects under real-world conditions with access to novel data sources. The Canton has confirmed this is explicitly not a regulatory sandbox: it operates entirely within existing law and grants no temporary relaxation of, or deviation from, existing legal requirements. The programme has been extended through 2029 as a non-exemptive test environment providing guidance and data access for AI pilots without relaxing any existing legal requirement.

Treaty track: Switzerland signed the Council of Europe Framework Convention on Artificial Intelligence, Human Rights, Democracy and the Rule of Law on 27 March 2025 but has not ratified it; a public consultation on implementing legislation is expected only at the end of 2026.