I make companies AI-ready — in real projects, not on slides.

Selected projects for: Coop · Betty Bossi · Flughafen Zürich · Rivella · SOS-Kinderdorf · Hine Engineering

These six questions come up in every project and every training session — usually in this order. The answers are the same ones I give in the room.

Does the EU AI Act apply to us as a Swiss company?

Switzerland has no AI act of its own along EU lines. What applies is existing law — the FADP, the Unfair Competition Act (UWG), the Code of Obligations (OR) and so on — all of it technology-neutral. The AI Act, however, has extraterritorial effect: as soon as your output or your customers are in the EU, its obligations apply to you. For Swiss SMEs doing business in the DACH region, the pragmatic answer is therefore usually this: working to AI Act standards means one standard for all markets instead of two sets of rules side by side.

Are we even allowed to use AI — what about data protection?

Yes, with rules. Personal data stays out or gets anonymised; for critical use, take an enterprise licence where data processing is governed by contract. Then it is compliant. The question is rarely whether you are allowed — it is whether your people know what is allowed.

Our people have been using ChatGPT privately for a long time. Should we ban it?

Ban it and they will do it in secret — and then company data leaves through private accounts without anyone knowing. The sentence I hear most often, translated from the German: “But our apprentice has ChatGPT on his phone anyway.” That is exactly why you need rules instead of a ban: approved tools, clear no-gos, a policy you can read in five minutes. Controlled use with guardrails beats uncontrolled use without them, every time.

Who owns the AI output?

Copyright protects human creation. Pure AI output without a human creative contribution is generally not protected in Switzerland or in the EU — you may use it, but you rarely hold exclusive rights to it. Your real risk lies on the other side anyway: third-party rights in the output, meaning trademarks, protected works, image rights. The practical rules are simple: edit the output substantially, check trademark and image rights, and never feed in someone else's protected works without asking.

Do we have to disclose when AI was involved?

Legally, in Switzerland: not as a general rule. Practically: with images always, with text at the latest when someone asks. When in doubt, declare it — transparency protects more than it costs.

Which AI tool should we buy?

The tool question is the last question, not the first. Four things belong before it: which use case are we solving, and which metric do we measure it against? Which data flows into it — and what does the FADP have to say about that? How does it integrate with existing systems? And who reviews and signs off? Once that stands, the choice of tool is usually trivial. To get started, five to ten licences of ChatGPT Plus or Claude Pro are enough; after three months you evaluate with real usage data instead of vendor promises.

On the legal points: that is the map, not the legal opinion. The binding answer for your specific case comes from your legal department — I tell you which questions you need to put to them.