Home Insights About Free Consultation
DE | EN
Home · Insights · Compliance & Setup
Compliance & Setup

The EU AI Act in 2026 — what Swiss companies need to build into their setup

If you serve European clients from Switzerland, the AI Act belongs in your setup — not in next year's compliance review.

May 2026·7 min read·Vladimir Lysow
EU AI Act 2026
7 %
of global turnover — maximum penalty for AI Act violations
EU AI Act 2024
Aug 2026
binding deadline for high-risk AI systems
European Commission
4
risk classes — decisive for how you structure your business
EU AI Act 2024

The EU AI Act is not an EU problem. It is a problem for every Swiss company that uses AI and has clients in Europe. For founders and owners, this means one thing: compliance belongs in the setup, not in a retrofit project two years from now. Companies built with the AI Act in mind from day one avoid the structural cost. Those who don't will pay twice — in fines and in restructuring.

When the EU AI Act actually applies to a Swiss company

The Act has extraterritorial effect. It applies whenever the output of an AI system is used in the EU — regardless of where the operator is based. In practice, this captures almost every Swiss provider with European clients:

If you have no EU footprint — pure Swiss market, Swiss clients — you remain under the lighter Swiss regime. But most ambitious Swiss companies grow into the European market sooner or later. The question is not whether the AI Act becomes relevant, but when.

The four risk classes

The Act takes a risk-based approach. Depending on use, your AI system falls into one of four categories:

Class 1Unacceptable risk — prohibited since February 2025

  • Social scoring by public authorities
  • Subliminal manipulation
  • Biometric mass surveillance in public spaces

Class 2High risk — strict obligations from August 2026

Applies to AI systems used in:

  • HR and recruiting
  • Credit assessment and lending
  • Critical infrastructure
  • Education and vocational training

These systems require certified risk management, comprehensive documentation, and mandatory human oversight.

Class 3Limited risk — transparency obligations

Chatbots must be identifiable as AI. Deepfakes and AI-generated content must be labelled. Most modern SMEs with a chatbot fall under this class.

Class 4Minimal risk — no specific obligations

AI spam filters, AI in video games, basic recommenders.

"Building compliance after market entry costs twice — once in fines, once in structural rework. Building it into the setup turns it into a competitive advantage."

— UniExe – Suisse

Setup decisions that make the AI Act easier

From our advisory work with companies serving the EU, three structural choices have the most impact on AI Act readiness:

1. A holding structure with clear market separation

A Swiss holding with separate subsidiaries for the Swiss market and the EU market lets you isolate AI Act exposure cleanly. Swiss products keep more flexibility; EU products meet the stricter standard from day one.

2. An EU entity or authorised representative

For high-risk systems, the Act requires an EU establishment or an authorised EU representative. Planning this early lets you choose the location strategically (Dublin, Amsterdam, Munich). Late planning means improvising under deadline pressure.

3. Documentation and data architecture

High-risk AI requires technical documentation, risk assessments and demonstrable human oversight. Companies that build this into their workflow from day one — even a simple internal AI register — meet the requirements without additional effort. Those who try to reconstruct it later are working from emails.

Switzerland's own position remains an advantage

According to the Federal Council's 2025 report on AI regulation, Switzerland follows a technology-neutral, principles-based approach. FINMA defines four core expectations for AI in financial services: governance, transparency, robustness, and non-discrimination.

Substantively this aligns closely with the EU AI Act — but with significantly less bureaucratic overhead. Swiss companies that build compliance into their structure now are doubly positioned: lean for the Swiss market, fully EU-compliant for European clients.

Your setup checklist for August 2026

Step 1 — Market mapping

Which EU countries are your clients in today, or potentially tomorrow? Without this, no meaningful compliance strategy is possible.

Step 2 — Build an AI inventory

List every AI tool your company uses — including embedded AI in third-party software (CRM, accounting, marketing).

Step 3 — Classify the risk

For each system: is it used in HR decisions, credit assessment, or critical processes? High risk means strict obligations.

Step 4 — Review the structure

Do you need an EU subsidiary, an authorised representative, or a holding rearrangement? These decisions take time. They are not last-minute decisions.

Step 5 — Check data protection

Are client data flowing into public AI models? Swiss banking secrecy (Article 47 Banking Act) and the revised data protection law (FADP) set clear limits.

Step 6 — Build documentation as you go

High-risk systems need technical documentation, risk analyses and evidence of human oversight — captured continuously, not reconstructed in retrospect.

Build a setup that's AI Act-compliant — from day one.

UniExe structures companies that combine Swiss advantages with EU compliance. We assess your AI use, review the market fit, and align your holding so compliance is built in, not bolted on. First 45 minutes without fee.

Book an introductory call →