¶ Demand 8: Protect Innovation
Regulation must not become a moat for large corporations.
¶ Core Demand
Proportional requirements for SMEs, regulatory sandboxes and clear deadlines. Switzerland has the opportunity to develop a leaner regulatory approach than the EU -- and to demonstrate how it can be done better.
¶ The Problem: Regulation as a Market Entry Barrier
¶ The EU AI Act as a Warning
The EU AI Act is the world's first comprehensive AI law. It classifies AI systems by risk level and requires extensive documentation, conformity assessments and ongoing monitoring for high-risk systems [1].
The intention is correct. The impact could be fatal:
- Compliance costs: Estimates range from 200,000 to 400,000 euros per AI system for the initial conformity assessment. For a start-up with ten employees, this can be existentially threatening [2].
- Asymmetric burden: Google, Microsoft and Meta can afford entire compliance departments. A Zurich start-up cannot.
- Legal uncertainty: Many terms in the AI Act are vaguely defined. What exactly is a "high-risk system"? When is an AI system "sufficiently transparent"? Interpretation will take years -- years in which European start-ups hesitate while American and Chinese giants continue to invest.
¶ The Numbers
The EU invests around 1 billion euros per year in AI research. The USA invests over 60 billion dollars (private and public). China invests comparable sums [3]. If Europe channels its limited resources into compliance rather than innovation, it will become a regulated desert with imported technology.
¶ The Swiss Way: Proportional and Pragmatic
¶ Proportional Requirements
Not every AI system needs the same regulation. A chatbot for a bakery is not a medical diagnostic system. Switzerland should introduce a tiered model:
| Risk Level | Examples | Requirements |
|---|---|---|
| Minimal | Chatbots, recommendation systems, spam filters | Labelling as AI, no further obligation |
| Low | AI in marketing, logistics, customer service | Documentation, complaints mechanism |
| High | Medicine, justice, credit decisions, law enforcement | Full conformity, independent audit |
| Prohibited | Biometric mass surveillance, social scoring | Absolute ban |
¶ Regulatory Sandboxes
Regulatory sandboxes are protected environments in which start-ups and research institutions can test AI systems without having to meet all regulatory requirements immediately. The Swiss Financial Market Supervisory Authority FINMA has operated a fintech sandbox since 2019 -- the same model can be applied to AI [4].
Advantages:
- Start-ups can test and iterate more quickly
- Regulators learn from practice before they regulate
- Switzerland attracts international AI companies
¶ Clear Deadlines
Regulatory procedures must be time-limited. If an authority does not decide within 90 days, the approval is deemed granted (deemed approval). This prevents innovation from dying in bureaucratic queues.
¶ Swiss Competitive Advantage
Switzerland has held a top position in the Global Innovation Index for years [5]. This lead is not a law of nature -- it must be defended. The combination of excellent research (ETH, EPFL), strong patent protection, a stable legal order and pragmatic regulation is a competitive advantage that no other country offers in this form.
AI regulation that stifles innovation would destroy this advantage. AI regulation that addresses risks while maintaining freedom would strengthen it.
¶ Sources
[1] EU AI Act, Regulation EU 2024/1689.
[2] AlgorithmWatch, Compliance Costs of the EU AI Act: Initial Estimates, 2024.
[3] Stanford HAI, AI Index Report 2024: Global AI Investment Trends.