2027 is already on the agenda.

A Switzerland that regulates in small steps

Switzerland has no cross-cutting law on artificial intelligence. AI is not developing in a legal vacuum for all that. The Data Protection Act applies to the processing of personal data by AI systems, as the Federal Commissioner pointed out as early as 2023, and sector regulations, from finance to health, continue to apply.

The framework is nonetheless going to change. On 12 February 2025, the Federal Council decided to ratify the Council of Europe Framework Convention on artificial intelligence, which Switzerland signed on 27 March 2025. Rather than an equivalent of the European AI Act, it favours targeted adjustments to existing law. A preliminary draft for consultation, prepared by the FDJP with DETEC and the FDFA, is now expected for early 2027, according to a communication from the Confederation published in September 2026.

For a Swiss company, this pace may seem comfortable. It is only comfortable for strictly domestic activities.

Europe, neighbour and market

The European AI regulation applies to Swiss companies that place AI systems or models on the EU market. It can also concern them when the output produced by a system is used in the EU.

Its timetable has been adjusted. Prohibited practices have been banned since February 2025, and the rules on general-purpose AI models have applied since August 2025. The digital omnibus on AI, published in the Official Journal in July 2026, postponed the obligations for high-risk systems: to 2 December 2027 for standalone systems (recruitment, credit, education, critical infrastructure) and to 2 August 2028 for AI built into regulated products. The postponement affects the timetable, not the content of these obligations.

The transparency obligations, for their part, have not been pushed back. Since 2 August 2026, a person must be informed when interacting with an AI system such as a chatbot, and deepfakes must be disclosed. Technical marking of synthetic content by providers will follow on 2 December 2026. A conversational assistant installed on a Swiss website addressing a European clientele falls directly within this scope.

Sovereignty: from intention to architecture choices

Digital sovereignty has taken concrete form in Switzerland. In September 2025, EPFL, ETH Zurich and the Swiss National Supercomputing Centre (CSCS) released Apertus, a fully open language model whose weights, training data and methodology are documented. Available in 8 and 70 billion parameter versions, it was trained on more than 1,000 languages, including Swiss German and Romansh, with around 40% non-English data. Swisscom offers it on its infrastructure, and a version 1.5, able to analyse images and sound, was released in 2026.

Apertus was not designed as a consumer assistant meant to compete with the most advanced commercial models. It shows that an open, auditable alternative that can be hosted in Switzerland exists for certain uses. For an organisation, sovereignty thus becomes a series of architecture choices rather than an abstract principle: where data are processed, which model for which use, what dependence on a provider is acceptable, what reversibility if conditions change.

The calendar that matters

Several deadlines converge over the next fifteen months. Not all of them concern every organisation, but their proximity justifies an overall reading. They add to those already in force: the European Accessibility Act since June 2025, the Swiss obligation to report cyberattacks on critical infrastructure since April 2025, and the reporting of exploited vulnerabilities under the Cyber Resilience Act since September 2026.

What an organisation can do now

Nothing in this calendar calls for haste. Everything justifies preparing the ground.

• List the uses of AI in the organisation, including those that escape IT: online tools, extensions, assistants built into office suites.

• Adopt a simple usage policy: authorised tools, prohibited data, required human validation.

• Clearly flag conversational interfaces and generated content.

• Map data flows to AI providers and check the corresponding contracts.

• Identify EU-facing activities, which already fall under European law.

This article presents a general framework and does not constitute legal advice. The dates cited are those known at the time of publication.

Frequently asked questions

Will Switzerland have a law comparable to the AI Act?

The Federal Council has chosen another route: ratify the Council of Europe Convention on AI and adjust existing law in a targeted way. A preliminary draft is expected for consultation in early 2027.

Is a Swiss company subject to the AI Act?

Yes if it places AI systems or models on the EU market, and potentially when the output of its systems is used in the EU.

Have the AI Act obligations been postponed?

Only in part. The digital omnibus postpones the high-risk obligations to December 2027 and August 2028, but the prohibitions, the rules on general-purpose models and the transparency obligations already apply.