AI Governance in Switzerland
Switzerland is not directly subject to the EU AI Act. But concluding that its content has no relevance for Swiss companies would be short-sighted. Anyone active in the EU, running AI systems there, or providing services to European clients will face its requirements. Many global technology vendors already align with the EU AI Act today.
Something else matters more. The questions the EU AI Act raises are not legal technicalities. They are fundamental governance questions, and they concern every well-run company, including in Switzerland.
AI shapes decisions, even without formal decision-making power
Artificial intelligence is already operational in many companies. It prioritises customer requests, supports credit and risk decisions, screens job applications, and flags anomalies in processes. This often happens gradually, in a decentralised way, embedded in standard software or cloud services.
Formally, AI makes no strategic decisions. In practice, though, it shapes the space in which decisions get made. It determines which information is visible, which cases get priority, and which risks appear relevant. AI does not decide in the legal sense. It still has a say.
Whether a decision is automated, data-driven, or prepared by an external system makes no difference. Responsibility stays with the company and its leadership bodies.
AI governance is therefore not an IT topic. It is a matter of steering and liability.
Switzerland’s starting point: a liberal, innovation-friendly approach
Switzerland is taking an open, liberal, innovation-friendly approach. It relies mainly on self-regulation, while also pursuing topic-specific and sector-specific regulatory work. In the long run, it will probably not escape the EU’s much stricter framework (the EU AI Act 2024) entirely, for instance because of potential trade barriers or restricted market access.
In 2025, the Federal Council instructed the Federal Department of Justice and Police to prepare a consultation draft on new AI rules by the end of 2026. Switzerland has also signed the Council of Europe’s AI Convention, which sets human-rights and rule-of-law guardrails for the use of AI.
For boards, this means something concrete: governance expectations arise even without direct legal obligations.
Technical traceability has its limits
Transparency is a core element of classic governance. Decisions should be traceable, explainable and reviewable, what is known as explainable AI (XAI). With modern AI systems, though, this is only possible to a limited degree.
Current research on explainable AI and technical AI governance is clear. For complex models, full explainability and unambiguous causal attribution are not reliably achievable by technical means.
Explanatory models offer approximations, not guarantees, and different methods sometimes produce conflicting results. This limitation is not an implementation flaw. It is a structural feature of modern AI.
For governance, this is a critical insight. It forces a change in perspective.
Why good AI governance is not a brake on innovation
Governance is often blamed for slowing innovation down. That view falls short. In practice, missing governance holds innovation back far more than clear guardrails do.
Without defined responsibilities, accepted risk limits and clear decision rules, uncertainty takes hold. Projects get delayed, decisions get postponed or revised later. With AI in particular, this often leads to a creeping standstill: systems are technically possible but never cleared organisationally.
Good AI governance, by contrast, builds decision-making capacity. It allows organisations to take on risk deliberately, instead of pushing it aside implicitly. Teams know where AI may be used, which requirements apply, and when to escalate. That speeds up innovation. Governance is the reason, not the obstacle.
For the board, this is the key point: governance is not a control instrument. It is a leadership instrument.
What the board can do in practice
Because not everything can be measured or explained technically, governance becomes a leadership task. The board does not need to understand AI. It needs to decide where limited traceability is acceptable, and where it is not.
The central step is this: stop letting risk in by default. Accept it, or rule it out, explicitly. Wherever AI is used, risk follows, regardless of whether it was formally approved. It is the board’s task to define in which decision areas the use of AI is justifiable and where clear limits apply.
This is not about model quality. It is about limits of use. For decisions with legal, financial or reputational weight in particular, it must be defined whether, and in what form, human control remains mandatory. This prevents AI from gradually taking over tasks that were never consciously approved.
Clear ownership matters just as much. Every AI application needs to sit with an accountable business unit. Responsibility cannot stay with IT or with the vendor. When technical transparency is limited, organisational clarity becomes decisive.
Reporting to the board must change too. Detailed technical reports do not create steering capacity. What matters are questions of impact and consequence: where does AI influence decisions today? What overrides or incidents occurred? Which risks would we no longer accept today?
Conclusion
For Switzerland, the EU AI Act is not law today, but it is a reference framework. National and international developments point the same way: AI governance is becoming part of modern corporate leadership.
The decisive question for the board and the executive team is not whether AI is fully explainable. It is where its use is justifiable, and where it is not.
Where technical traceability ends, leadership responsibility begins.
If AI governance is not yet an explicit board topic, it is worth a look. I support leadership bodies in building decision-making capacity around AI, beyond the technical detail debates. Feel free to get in touch for an initial conversation.