AI and Data Protection in Switzerland: What the FADP Requires

Key Takeaways

• Switzerland has no AI-specific law yet, but the existing Data Protection Act already applies directly to AI.
• As soon as information about people goes into an AI tool, the FADP principles apply: a clear purpose, only the data you need, accurate results and secure handling.
• Most AI providers act as your processor, so you need a contract with them and the responsibility stays with you.
• Tell the people concerned openly, and leave decisions about individuals to a human.
• Where a use could pose a high risk to people, the FADP requires a data protection impact assessment before you start.

Switzerland's Federal Act on Data Protection (FADP, known in German as DSG) applies directly to AI whenever a tool processes information about people. It doesn't need a special AI law to do so, and Switzerland doesn't have one yet. In practice the FADP asks four things of your business: know which personal data goes into which tool, have a proper contract with the provider, inform the people concerned, and keep important decisions about individuals with a human.

This guide explains the basics for small and mid-size businesses. It is not legal advice and reflects the position as of October 2026.

Does data protection law apply to AI at all?

Yes. The Federal Data Protection and Information Commissioner (FDPIC) has made clear that the FADP is worded in a technology-neutral way and therefore applies directly to AI-supported data processing. You'll find the Commissioner's guidance on the FDPIC page on AI and data protection and the law itself on fedlex.admin.ch. The revised FADP has been in force since 1 September 2023.

What matters is personal data, meaning any information relating to an identified or identifiable natural person. AI as a technology isn't covered by the FADP. But as soon as a tool processes personal data, the same rules apply as for any other software. Since 2023 the FADP no longer protects data about companies, only about people, so the contact person at a client firm is covered even though the firm itself isn't.

When are you processing personal data with AI?

More often than you might think. Under the FADP, processing means any handling of personal data, from collecting and storing it to using, disclosing and deleting it. Typical office examples:

• An AI tool summarises or answers a customer email that includes a name, an address and a request.
• An assistant sorts job applications or summarises CVs.
• A tool records a meeting and writes up the minutes.
• Your team searches client, tenant or project files with AI.
• A document contains health information, such as a doctor's note. That counts as sensitive personal data, as do details about religion, criminal proceedings or social assistance.

No personal data is involved when you draft a general text about your opening hours or ask for help with a spreadsheet formula. For tasks like these the FADP doesn't come into it.

Which principles do you have to follow?

The same ones as for any other processing. Article 6 FADP requires you to process personal data lawfully, in good faith and proportionately, only for a purpose the person can recognise, and for no longer than necessary. In everyday AI use, that means:

• Purpose: use data only for what you received it for, or for something compatible with that. Answering tenant inquiries with AI fits. Analysing the same emails to rate tenants is a new purpose.
• Proportionality: put in only as much data as the task needs. Often the text works just as well without the name, address and customer number.
• Accuracy: whoever processes data has to make sure it's correct. An AI summary about a person gets checked before it goes into a file. We explain why AI invents things in why AI makes things up.
• Security: protect the data with technical and organisational measures that match the risk, such as business accounts instead of private ones and clear access rights.
• Privacy by default: choose settings that process and keep only what's needed.

What applies when an AI provider processes your data?

If a provider processes data on your behalf, the FADP treats it as your processor, and the responsibility stays with you. That's the case with most AI tools used in an office. Article 9 sets clear conditions. The provider may only process the data the way you yourself would be allowed to. You have to satisfy yourself that it can guarantee data security. It may bring in further subcontractors only with your prior approval. And no legal or contractual duty of confidentiality may stand in the way, which matters for lawyers, notaries and auditors bound by professional secrecy. We cover that in AI in law firms.

In practice this means a business contract with a data processing agreement rather than a free private account. Also find out whether the provider uses your inputs for its own purposes, such as improving its models. If the data is stored abroad, or someone abroad can access it, the rules on disclosure abroad apply as well. Where your data goes explains what to look for.

What do you have to tell the people concerned?

When you collect personal data, you have to inform the people concerned appropriately: who is responsible, what the data is used for, who receives it and which countries it goes to. If you use AI tools on customer data, this belongs in your privacy notice, with the categories of recipients and the countries. The FDPIC also stresses that anyone chatting with a bot has the right to know they're dealing with a machine and whether their input will be reused. Manipulated images or voices of identifiable people must always be clearly recognisable as such.

Article 21 deserves special attention. If a decision is based solely on automated processing, and it has legal consequences for the person or significantly affects them, you have to tell them. They can then give their point of view and ask for a human to review the decision. The exceptions are narrow, for example when the decision is directly linked to a contract with the person and gives them what they asked for, or when they have explicitly agreed to it. The simplest way to stay on the safe side: when you turn down an applicant, a prospective tenant or a customer, a person makes that call.

When do you need a data protection impact assessment?

Whenever processing may pose a high risk to the personality or fundamental rights of the people concerned. The law names new technologies as a possible source of such risk and gives two examples: processing sensitive personal data on a large scale, and systematically monitoring large public areas. The assessment describes the planned processing, evaluates the risks and records the measures you take to reduce them. If the risk remains high despite those measures, you ask the FDPIC for its opinion beforehand, or consult your data protection advisor if you have appointed one.

For many office uses, where someone reviews every draft and no sensitive data is involved, the risk is likely to be low. It's different if you analyse health data at scale, rate people or automate decisions. Assess each use on its own and write the result down. As for the record of processing activities: businesses with fewer than 250 employees only need one if they process sensitive data on a large scale or carry out high-risk profiling.

What happens if you break the rules?

The FADP provides for fines of up to CHF 250,000, for example for intentionally breaching the duty to inform, sending data abroad without adequate protection, or handing data to a processor without the conditions being met. These fines are aimed first at the responsible individuals in the business, not at the company. On top of that come the damage to your customers' trust and possible civil claims by the people affected.

Five steps to a clean set-up

1. Take stock.Write down which AI tools your team uses today, private accounts included, and what data goes into them.
2. Check the contracts.Move to business accounts with a data processing agreement, and clarify where the data is held, which subcontractors are involved and how your inputs are used.
3. Set the rules.Write down which data may go into which tool, with examples from your own work. Our model outline for an AI policy gives you a starting point.
4. Be open about it.Update your privacy notice and make sure chatbots identify themselves as such.
5. Let people decide.Build a review step into anything that affects individuals, and carry out an impact assessment first where the risk is high.

An example: a property management firm wants AI to pre-sort incoming tenant inquiries and draft replies. It picks a tool with a data processing agreement and data held in Switzerland or the EU, updates its privacy notice and decides that the team reads every draft before it goes out. Sensitive cases such as notices to quit or disputes go straight to a person.

More is on its way in Switzerland: the Federal Council has commissioned a consultation draft of rules on AI by the end of 2026, covering transparency and data protection among other things. How this relates to the EU's rules is explained in the EU AI Act for Swiss companies. If you'd like to go through your own use of AI, get in touch.