Skip to content

Basics

EU AI Act: what applies to SMEs in Switzerland and Liechtenstein?

By Dominik Natter · Last updated on 3 November 2026 · 8 min read

In short

The EU AI Act applies to SMEs in Switzerland and Liechtenstein as soon as they offer AI systems in the EU or the output of their AI is used in the EU. Most SMEs are deployers: they must promote AI literacy among their staff and, since 2 August 2026, disclose certain AI-generated content. Following the Digital Omnibus, the strict high-risk obligations, for example for selecting job applicants, only apply from 2 December 2027. In Liechtenstein, incorporation into EEA law is not yet complete, and Switzerland is not adopting the AI Act but making targeted changes to its own laws.

When does the EU AI Act apply to an SME in Switzerland?

The AI Act also applies to companies based outside the EU as soon as their AI has an effect in the EU. Article 2 of the regulation covers providers that place AI systems on the EU market. It also covers providers and deployers in third countries where the output of their AI system is used in the EU. The same principle applies under the Cyber Resilience Act to software and connected devices offered in the EU. Read more in our guide "Cyber Resilience Act: what applies to software from SMEs in Switzerland and Liechtenstein?".

For a Swiss SME, this means: a chatbot on your website used by customers from Germany or Austria is covered. So are AI-generated images in a campaign for the EU market, or AI-assisted pre-screening of applications for a branch in the EU. If you only use AI internally and for Swiss customers, the AI Act generally does not affect you directly.

Transparency obligations under Article 50 apply since this date, not postponed
2 Aug 2026
High-risk obligations under Annex III apply from this date, for example in HR
2 Dec 2027
or 7 percent of annual turnover: maximum fine for prohibited practices
EUR 35m
for SMEs as deployers without a high-risk application, no certification
5 points

Switzerland has no AI law modelled on the EU. On 12 February 2025, the Federal Council decided to ratify the Council of Europe's AI Convention and to adapt Swiss law sector by sector wherever possible. Switzerland signed the Convention on 27 March 2025. According to OFCOM, the consultation on the necessary legislative changes is planned for spring 2027. Until then, AI is governed mainly by the Data Protection Act and other existing law.

Does the AI Act apply in Liechtenstein?

Not directly, yet. As an EEA member, Liechtenstein will only apply the AI Act once it has been incorporated into the EEA Agreement. According to EFTA, Iceland, Liechtenstein and Norway are still reviewing the regulation for incorporation. As of November 2026, there is no decision of the EEA Joint Committee.

Liechtenstein will then need an implementing act that designates the competent authorities and sets out penalties. The Office for Digital Innovation of the national administration is drafting it. It has not given a date for entry into force.

In practice, this changes little. Liechtenstein SMEs whose AI output is used in the EU already fall under the AI Act through the market location principle, just like Swiss companies. The Office for Digital Innovation recommends preparing now. Once incorporation is complete, the AI Act will also apply in Liechtenstein to purely domestic cases.

Provider or deployer: which role does your SME have?

Most SMEs are deployers. A deployer is anyone who uses an AI system under their own authority in a professional context. If you use Microsoft Copilot, ChatGPT or the invoice recognition built into your accounting software, you are a deployer.

AI Act deadlines 2026: what the Digital Omnibus has changed

The Digital Omnibus postpones the high-risk obligations, not the transparency obligations. The Council and Parliament reached a political agreement on 7 May 2026. Regulation (EU) 2026/1744 was adopted on 8 July 2026, published in the Official Journal on 24 July 2026 and entered into force on 27 July 2026.

The deadlines for the prohibitions, AI literacy, providers of general-purpose AI models and the transparency obligations under Article 50 remain unchanged. The timeline shows all deadlines and the main changes for SMEs.

EU AI Act deadlines and status in Switzerland and Liechtenstein

  1. 1 August 2024

    AI Act entered into force

    Status: in effect

  2. 2 February 2025

    Prohibited practices and AI literacy under Article 4

    Status: in effect, Article 4 relaxed since 27 July 2026. Measures are required, not a specific level of competence per person.

  3. 2 August 2025

    Obligations for providers of general-purpose AI models

    Status: in effect, applies to model providers, not users

  4. 2 August 2026

    Transparency obligations under Article 50 for chatbots, deepfakes and AI content

    Status: in effect, not postponed

  5. 2 December 2026

    Marking for generative AI already on the market before 2 August 2026, and new prohibitions

    Status: adopted with Regulation (EU) 2026/1744. Providers of such systems have until then to add machine-readable marking. AI systems that generate intimate images of real people without consent or child sexual abuse material become prohibited.

  6. 2 December 2027

    High-risk systems under Annex III, such as HR and credit

    Status: postponed from 2 August 2026, adopted. SMEs that provide high-risk systems themselves may prepare the technical documentation in a simplified form.

  7. 2 August 2028

    High-risk AI in regulated products under Annex I

    Status: postponed from 2 August 2027, adopted. Covers AI in products such as medical devices or toys.

  8. Liechtenstein

    Incorporation into the EEA Agreement and implementing act

    Status: pending, no date set

  9. Switzerland

    Consultation on implementing the Council of Europe AI Convention

    Status: planned for spring 2027

What an SME actually needs to do now

For an SME that uses AI as a deployer without a high-risk application, there are five points: AI literacy under Article 4, recognisable chatbots, disclosed AI content, an inventory of AI use, and data protection under the Swiss Data Protection Act in Switzerland and the GDPR in Liechtenstein and the EU. None of them requires certification or notifying an authority. The checklist for the next 90 days further down turns these five points into concrete steps.

People chatting with an AI system must be told so, unless it is obvious. The obligation lies with the provider. As a deployer, check whether your tool shows the notice and add it yourself if not.

Our guide "AI automation in SMEs" explains in detail how to build AI workflows with human approval, an audit trail and a data protection concept.

High risk and fines: where it gets serious

High risk concerns a small number of clearly defined areas of use. From 2 December 2027, deployers of such systems must, among other things, ensure human oversight by trained staff, keep logs and inform employees and affected persons.

Typical high-risk cases in an SME:

  • HR: AI that screens or filters applications, evaluates candidates or influences decisions on promotion, dismissal and performance reviews.
  • Credit: AI that assesses the creditworthiness of natural persons, such as private customers. Systems used to detect fraud are exempt.
  • Change of purpose: a general-purpose language model that you use for one of these purposes.

What you do not need to do

Most AI automation in everyday SME work is not subject to high-risk obligations. Invoice recognition, email sorting and drafting quotes are not on the list of high-risk areas in Annex III. They do not assess people; they read documents and make suggestions. A person checks and approves.

The labelling of AI text does not apply here either. A quote or a reply to a customer is not a publication on matters of public interest. Machine-readable marking of AI-generated content is the model provider's job, not yours.

Still, review each workflow individually. As soon as it assesses employees or job applicants, the classification changes.

  • No conformity assessment, no CE marking and no registration in the EU database for systems that are not high-risk.
  • No AI officer: the AI Act does not require this role.
  • No certificate for AI literacy: Article 4 requires neither a specific level nor an exam.
  • No new software just because of the AI Act: spell checkers, translation tools and spam filters are not high-risk applications.

Checklist for the next 90 days

These steps cover the obligations that apply today and show early on whether a high-risk case is involved. For an SME, this is a manageable effort, not a major project.

  1. 01

    Days 1 to 30: create an inventory

    Record all AI tools, including AI features in existing software, with purpose, data, vendor and responsible person. Not explicitly required by the AI Act, but needed to assess your role and the risk.

  2. 02

    Days 1 to 30: clarify your role

    For each tool, clarify whether its output is used in the EU and whether you are a deployer or a provider.

  3. 03

    Days 31 to 60: write an AI policy

    Write an AI policy of two to three pages covering permitted tools, data that must not be entered and human approval.

  4. 04

    Days 31 to 60: train staff

    AI literacy under Article 4: give staff who work with AI a short training on benefits, limits and risks, and record each session with date and content.

  5. 05

    Days 61 to 90: label AI content

    Review chatbots and publicly visible AI content. Label realistic AI images, videos and audio. Label AI text only if it covers matters of public interest and no person has reviewed it editorially.

  6. 06

    Days 61 to 90: review contracts

    Review contracts with AI vendors: purpose, place of processing, no training on your data, data processing agreements under the FADP and GDPR. According to the FDPIC, the FADP applies directly to AI.

  7. 07

    Days 61 to 90: plan for high risk

    Flag potential high-risk cases such as AI in recruitment and plan the necessary work ahead of 2 December 2027.

Where this information comes from

EU regulations

The basis is Regulation (EU) 2024/1689 on artificial intelligence as published in the Official Journal of the EU, in particular Articles 2, 3, 4, 25, 26, 50, 99 and 113 and Annex III. The amendments come from Regulation (EU) 2026/1744 of 8 July 2026, the Digital Omnibus on AI.

Liechtenstein and Switzerland

On Liechtenstein: EFTA, entry for Regulation (EU) 2024/1689 in the EEA-Lex database, and information from the Office for Digital Innovation of the Liechtenstein national administration. On Switzerland: the Federal Office of Communications OFCOM, artificial intelligence page as of 28 October 2026, and the Federal Data Protection and Information Commissioner FDPIC on the application of the Data Protection Act to AI.

Scope of this guide

This guide is for orientation only and does not constitute legal advice. Only a specialist who knows your case can assess it with certainty.

As of: November 2026.

1 August 2024AI Act entered into forceStatus: in effect
2 February 2025Prohibited practices and AI literacy under Article 4Status: in effect, Article 4 relaxed since 27 July 2026
2 August 2025Obligations for providers of general-purpose AI modelsStatus: in effect, applies to model providers, not users
2 August 2026Transparency obligations under Article 50 for chatbots, deepfakes and AI contentStatus: in effect, not postponed
2 December 2026Marking for generative AI already on the market before 2 August 2026, and new prohibitionsStatus: adopted with Regulation (EU) 2026/1744
2 December 2027High-risk systems under Annex III, such as HR and creditStatus: postponed from 2 August 2026, adopted
2 August 2028High-risk AI in regulated products under Annex IStatus: postponed from 2 August 2027, adopted
LiechtensteinIncorporation into the EEA Agreement and implementing actStatus: pending, no date set
SwitzerlandConsultation on implementing the Council of Europe AI ConventionStatus: planned for spring 2027

Frequently asked questions

Does the EU AI Act apply to Swiss companies?

Yes, if you offer AI systems in the EU or the output of your AI is used in the EU, for example a chatbot used by customers in Germany. This is set out in Article 2 of the regulation. If you only use AI internally and for Swiss customers, you are generally not directly affected. Switzerland itself is not adopting the AI Act; it is making targeted changes to its own laws instead.

Do SMEs have to train their staff because of the AI Act?

Yes, Article 4 of the AI Act requires measures to build the AI literacy of staff who work with AI. Since the Digital Omnibus of July 2026, no specific level of competence per person is required and no certificate is needed. For most SMEs, a short training session, an internal policy and a record of who was trained when are a proportionate approach.

Does AI-generated text have to be labelled?

Usually not. The AI Act only requires disclosure for AI-generated text that informs the public on matters of public interest and has not been reviewed by a person who takes editorial responsibility for it. Quotes, replies to customers or reviewed blog posts are not covered. Realistic AI images and videos that could pass as genuine, on the other hand, must be labelled.

Where does our data go when AI is involved?

We set that out in writing for each project. The default is processing in a European region, under contracts that rule out any use of your data for training. For sensitive data, self-hosted models are an option. Hosting and database are located in the EEA, and processing is designed around the GDPR and the Swiss Federal Act on Data Protection.

What is the discovery phase and why does it come first?

In two to three weeks, the discovery phase establishes what will be built, how it fits into your systems and what it will cost. You receive requirements, a solution design, a stage plan and a cost ceiling. The price is fixed, usually CHF 4,000 to 9,000. After that you are free to decide whether to go ahead, and with whom.

Matching services:AI automation

Let us talk about your case.

20 minutes are enough to find out whether a custom solution is worth it for you.