Swiss nFADP and generative AI: sovereignty, transfers and secrecy

In Switzerland, discretion is part of the country's economic identity. Private banks in Geneva, pharmaceutical headquarters in Basel, wealth managers in Zurich and Lugano: client trust rests on the certainty that their information stays where it belongs. Generative AI is shaking that certainty.

When an analyst pastes a client statement into ChatGPT, or a researcher summarises a protocol with an online assistant, the data often leaves Switzerland. This is no longer just a technical question. It touches data sovereignty, Swiss law and the professional secrecy rules that protect entire sectors.

The revised FADP already applies to AI

The revised Federal Act on Data Protection (nFADP) came into force on 1 September 2023. It was written to be technology neutral. The Federal Data Protection and Information Commissioner (FDPIC) made this point in November 2023, then again in an update published in May 2025: the law applies directly to AI supported data processing.

In practice, the FDPIC expects manufacturers, providers and users of AI systems to ensure:

• transparency about the purpose, functioning and data sources of the processing;

• that people know when they are dealing with a machine, and whether their input is used to train the system;

• respect for the right to object to automated processing and to ask for an automated individual decision to be reviewed by a human;

• a data protection impact assessment where the risk is high.

The FDPIC also names uses that are contrary to the law, such as comprehensive real time facial recognition or social scoring.

Transfers abroad: the real sovereignty issue

Most consumer AI assistants process data outside Switzerland. The nFADP strictly governs the disclosure of personal data abroad. The principle: a transfer is allowed to a country on the Federal Council's list, in Annex 1 of the Data Protection Ordinance. Otherwise, safeguards are needed, such as the standard contractual clauses recognised by the FDPIC, or an exception provided by law.

Since 15 September 2024, the United States has been on that list, but only for companies certified under the Swiss-US Data Privacy Framework. An AI tool whose provider is not certified, or whose hosting location is unknown, is therefore a problem. The law also requires people to be informed when their data goes abroad. We look at these issues in our article on sovereign AI for European CISOs.

Banking secrecy and professional secrecy

For some sectors, the nFADP is only the first layer. Banking secrecy is protected by Article 47 of the Banking Act, and breaching it is a criminal offence. Professional secrecy under Article 321 of the Swiss Criminal Code covers doctors, pharmacists and lawyers, among others. The Criminal Code also protects manufacturing and business secrets.

These duties do not vanish because a tool is convenient. Sending a bank client's name, a medical file or a still confidential formulation to an external AI service can expose the company, and sometimes the person who did it.

Banks and pharma: what supervisors look at

FINMA published its supervisory guidance 08/2024 on AI governance in December 2024. It expects institutions to keep an inventory of AI applications and to manage data and third party risks, and it stresses that responsibility for decisions stays with the institution, not with the tool or the provider.

In pharma, the stakes are slightly different but just as sensitive: clinical trial data, marketing authorisation files, patient health data, unpublished research. A leak can hit data protection, professional secrecy and the value of intellectual property all at once.

What are the penalties?

The nFADP provides for criminal fines of up to CHF 250,000. A Swiss particularity: they are aimed in principle at the responsible individuals, not the company. A data security breach that is likely to result in a high risk must be reported to the FDPIC as soon as possible. Managers therefore have a very personal interest in seeing AI rules applied.

Staying in control without banning AI

Banning ChatGPT solves nothing: teams switch to their phones or personal accounts, and the problem becomes invisible. A sounder approach combines four elements:

1. An inventory of the AI tools actually in use and the countries where they process data.

2. A clear policy separating public, internal, personal and secrecy protected data.

3. A control at the source that detects sensitive data before it leaves the workstation.

4. A record that can be shown to the FDPIC, FINMA or an auditor.

An AI DLP solution covers the last two points. It anonymises names, account numbers or identifiers before sending through format preserving anonymisation, warns or blocks, and logs every incident. For a gradual rollout, our 90-day rollout playbook offers a concrete method.

Key takeaways

In Switzerland, generative AI is first a question of sovereignty. The nFADP applies fully to AI, transfers abroad are regulated, and banking and professional secrecy add a criminal law layer. Companies that stay in control of what leaves their workstations can benefit from AI without giving up the discretion their reputation depends on.

Want to see how ZeusLock keeps your sensitive data in Switzerland while your teams use AI? Book a demo.