ChatGPT and the FADP: what Swiss businesses can do.

At a glance

• The FADP, in force since 1 September 2023, also applies to AI. The federal data protection commissioner (FDPIC) has been saying so since 2023.
• Before you put personal data into an AI tool, answer six questions: what data it is, where it goes, under what contract, who has been told, who decides and how much risk is involved.
• A US provider is fine if it is certified under the Swiss-U.S. Data Privacy Framework, or if contractual safeguards are in place.
• Fines go up to CHF 250,000 for wilful breaches, and as a rule they are paid by the person who made the decision, not by the company.

Take a letting agency in Lugano. It advertises a 3½-room flat and receives forty applications. To save time, an employee pastes the files into a free AI chat and asks for a ranking of the five best tenants. The files contain names, salaries, copies of ID documents and debt collection register extracts. The ranking arrives within a minute.

The FADP doesn’t forbid using AI to choose a tenant. But in that minute several things happened that the law does regulate: personal data left the agency, possibly for another country, with no contract with whoever received it and without the applicants knowing. Here we look at how the FADP and artificial intelligence fit together, and what to do before you start.

Switzerland’s AI law already exists: it’s the FADP

On 12 February 2025 the Federal Council decided how Switzerland will regulate AI. The country intends to ratify the Council of Europe’s Convention on Artificial Intelligence, which mainly concerns state actors. Where new laws are needed, they are to be sector-specific wherever possible, for example in healthcare and transport. Rules that apply to everyone will stay limited to core areas such as data protection. The draft for consultation is due by the end of 20262.

For a business, then, the rule that applies today is the FADP, the Federal Act on Data Protection1. The Federal Data Protection and Information Commissioner (FDPIC) has been saying so since November 2023, and said it again in May 2025.

“The Federal Act on Data Protection (FADP) is formulated in such a way as to apply to all types of technology and is therefore also directly applicable to the use of AI-supported data processing.”

The law comes into play as soon as AI processes personal data, meaning any information about an identified or identifiable person1: a name, an email address, an invoice made out to someone, a note about a customer. AI as such, the FDPIC points out, is not subject to the law. It becomes subject to it when it processes personal data4. A product description or the translation of a price list stays outside it.

Companies are paying attention. For 51% of those EY surveyed in May 2026, it is crucial that AI complies with Swiss or European data protection law and that data is processed in Switzerland or the EU5. The sample is mostly large companies, but a three-person office faces exactly the same questions.

Six questions before you put data into an AI

Before you paste a document into a chat, or connect an AI tool to your email, invoices or customer records, answer these questions. Each one has its own article in the FADP1.

Six questions, the FADP article and what to do

Question	FADP	What to do
Is personal data involved? What about sensitive data, such as health information?	Art. 5	Remove names and numbers where you can. Keep health data out of tools that aren’t built for it.
Where does the data physically go?	Art. 16	Ask the provider which country it is processed in. Outside Switzerland and the EU, check the safeguards.
Does the provider work for you under a contract?	Art. 9	A business account with a data processing agreement, not an employee’s personal account.
Do the people concerned know?	Art. 19	Update your privacy notice: purpose, recipients and, if data leaves Switzerland, the country.
Does the machine decide something important on its own?	Art. 21	Have a person decide. If not, tell people and let them ask for a review.
Is the risk high?	Art. 22	Carry out a data protection impact assessment before you start.

Two pieces of jargon. A “processor” is anyone who processes data on your behalf: the AI provider, the cloud software, the agency that builds your automation. A “data protection impact assessment” is a document that describes the planned processing, the risks to the people concerned and the measures that protect them (Art. 22)1. According to the FDPIC, the law presupposes one wherever the risk is high3.

A separate point if you are bound by professional secrecy, as doctors, lawyers and auditors are. The FADP only lets you entrust data to a processor if no statutory or contractual duty of confidentiality forbids it (Art. 9)1.

Data abroad and US providers

Many well-known AI services are American. The FADP allows personal data to be sent abroad to countries the Federal Council considers adequate. Anywhere else, you need safeguards, such as contractual clauses or standard data protection clauses recognised by the FDPIC (Art. 16)1.

For the United States there is a simpler route, on one condition. Since 15 September 2024, US companies certified under the Swiss-U.S. Data Privacy Framework have been considered adequate, and data can go to them without further safeguards6. Companies that aren’t certified don’t get the simple route.

Then there’s the privacy notice: if data goes abroad, people must be told which country (Art. 19)1. Processing can also stay in Switzerland or Europe: on our projects we decide this up front and put it in writing in the data sheet.

When the machine decides

Back to the tenant ranking. If the agency contacts the top five and turns down the other thirty-five without anyone reading their files, the machine has made the decision.

The FADP has an article for exactly this. If a decision is based solely on automated processing and has legal effects or significant consequences for the person, you must inform them. On request, the person can give their point of view and ask for the decision to be reviewed by a human being (Art. 21)1. There are exceptions, for example when the person has given their express consent.

AI can draw up the list.
A person says yes or no.

When it comes to rentals, the FDPIC keeps a close eye even without AI. For years it has been following up on application forms that ask for more data than necessary, and in its 2025/2026 report it says it has intervened with several property management companies9. The principle is set out in its guidance: collect only the information that is suitable for choosing a tenant on objective criteria8. Documents such as a copy of an ID are requested later, and only from the shortlisted applicants9.

The same applies when AI talks to your customers. On 28 January 2026 the FDPIC reiterated that users “must be able to recognise when they are dealing with AI, as well as how the data from their prompts may be used”4. An assistant in a chat, on WhatsApp or on the phone introduces itself as an AI from the very first sentence.

Who is liable for what

Penalties under the FADP are criminal and personal. Fines go up to CHF 250,000 and apply to private individuals who act wilfully1. The cases below are prosecuted only on complaint:

• failing to inform people about the collection of their data or about an automated decision (Art. 60);
• sending data abroad without meeting the conditions of Art. 16 (Art. 61);
• entrusting data to a processor without meeting the conditions of Art. 9 (Art. 61);
• failing to meet the minimum data security requirements (Art. 61);
• disclosing secret personal data learned in the course of your profession (Art. 62).

250,000

The company can be fined instead of the individual only if the fine does not exceed CHF 50,000 and identifying the person responsible would require disproportionate investigation (Art. 64). Prosecution is a matter for the cantons (Art. 65)1.

The FDPIC does not impose fines, but it can order processing to be adapted, suspended or stopped, data to be deleted, and the disclosure of data abroad to be suspended or prohibited (Art. 51)1. For doctors, lawyers, notaries, auditors and their assistants there is also the Criminal Code: anyone who discloses a professional secret faces, on complaint, a custodial sentence of up to three years or a monetary penalty (Art. 321)10.

That doesn’t mean every mistake ends up in court. But it pays to be able to show, in writing, what you decided and why. This article is not legal advice: for your own situation, ask your fiduciary or a lawyer.

The data sheet: one page before you start

Before we build an automation or business software with AI, we give the client a single page. It says which data leaves the business, which AI model reads it (by name), where it is kept and for how long. You’ll find it described on our AI Automation and AI Business Software pages.

A page like this already answers most of the six questions. It also helps with another obligation: the FADP requires a record of processing activities, with exemptions for companies with fewer than 250 employees where the risk is low (Art. 12)1. The data sheet isn’t the record, but it’s a good start.

If you’d like to see what a data sheet looks like, we’ll show you in the free 30-minute consultation, before anything gets built. Sometimes two written rules and a business account are all it takes, and that’s that.

Sources

1. Federal Act on Data Protection (FADP), SR 235.1, Arts. 5, 9, 12, 16, 19, 21, 22, 51, 60, 61, 62, 64 and 65. Fedlex, in force since 1 September 2023.
2. AI regulation: Federal Council to ratify Council of Europe Convention. Federal Council, 12 February 2025.
3. Update: Current data protection legislation is directly applicable to AI. FDPIC, 8 May 2025 (first published 9 November 2023).
4. Data Protection Day 2026: Uses of generative AI and data protection issues. FDPIC, 28 January 2026.
5. EY survey on artificial intelligence in Swiss companies (in German). 604 respondents, more than half of them at companies with over 1,000 employees. EY Switzerland, 27 May 2026.
6. Swiss-US Data Privacy Framework: Certified US companies offer adequate protection for personal data. Federal Council, 14 August 2024.
7. Data Privacy Framework List. U.S. Department of Commerce.
8. Forms for renting flats (in Italian). FDPIC.
9. 33rd Annual Report 2025/2026 (in German), campaign on the property market. FDPIC, 30 June 2026.
10. Swiss Criminal Code (SCC), SR 311.0, Art. 321. Fedlex.

Written by Alexandru Ciobanu. Sources checked on 14 July 2026. If you spot a figure that doesn’t add up, email us at info@digitam.ch.