ChatGPT and Data Protection in Switzerland: What Firms Must Know
What the revised FADP, professional secrecy and the CLOUD Act mean for ChatGPT and Copilot at work, what an enterprise plan solves, plus a sample staff policy.
Privacy & Law9 min read
A clerk receives a long, muddled email from a client. She pastes it into ChatGPT, attaches the client's tax documents as a PDF and asks for a draft reply. It takes two minutes and saves half an hour. Whether her employer knows about it is another matter.
This happens a lot. In the 2024 Work Trend Index by Microsoft and LinkedIn, 78 per cent of AI users surveyed said they bring their own AI tools to work. This article sets out what that means under Swiss law and what options a firm has. It is not legal advice.
The Data Protection Act already covers AI
The revised Federal Act on Data Protection (FADP, in German DSG) has applied since 1 September 2023. The Federal Data Protection and Information Commissioner (FDPIC, in German EDÖB) has stated that the FADP applies directly to AI-based data processing. No separate AI law is needed for that.
As soon as a prompt contains personal data, the usual principles apply: processing must be proportionate and tied to a purpose, the data must be adequately secured, and any disclosure abroad needs a legal basis. A client's name, address, social security (AHV) number, salary certificate or health details are all personal data. In its campaign on generative AI at work, the FDPIC warns that people typing into AI tools often reveal far more than they realise.
For the US, the Swiss-U.S. Data Privacy Framework has applied since 15 September 2024: personal data may be sent to certified US companies without additional safeguards. That covers the transfer itself. It does not answer whether the disclosure is allowed in the first place, for example under professional secrecy.
Professional secrecy and the duty of confidentiality
For many offices, criminal law sets the tighter limit.
Art. 321 of the Swiss Criminal Code (StGB) protects the professional secrecy of, among others, lawyers, notaries, auditors bound by statutory confidentiality, doctors and their auxiliaries. Under the prevailing legal view and the Swiss Bar Association's rules, IT and cloud providers can count as auxiliaries if they are carefully selected, contractually bound to confidentiality and supervised (overview for professional secrecy holders). A private ChatGPT account that an employee uses on her own initiative does not meet those conditions.
Art. 62 FADP reaches further. It makes it an offence to intentionally disclose secret personal data learned while practising a profession that requires knowledge of such data. The fine is up to CHF 250,000, prosecuted on complaint (commentary on Art. 62 FADP). This also covers professions outside Art. 321 StGB, such as fiduciaries, tax advisers or HR staff. The offence is committed by the individual, not the company.
On top of this come contractual confidentiality duties in engagement letters and the professional rules of industry associations.
Free, Business or Enterprise: what changes
OpenAI distinguishes between personal and business accounts.
On personal accounts (Free, Plus, Pro), conversations may be used to train future models. Users can switch this off under "Improve the model for everyone". A firm, however, cannot check whether every employee has done so.
For ChatGPT Business (formerly Team), Enterprise and the API, OpenAI does not use inputs and outputs for training by default. Business customers get a Data Processing Addendum (DPA), an admin console and central user management.
Since February 2025 OpenAI has offered data residency in Europe for Enterprise, Edu and the API. According to its help centre, the Europe region covers the EEA and Switzerland. Since January 2026, eligible Enterprise customers can also have the computation itself run in Europe.
Microsoft has announced that it will offer in-country processing for Microsoft 365 Copilot in Switzerland by the end of 2026. Check whether it is already active for your tenant before you decide.
What the contract solves and what it does not
An enterprise contract with a DPA and European data residency settles a fair amount: no training on your data, a contractual basis for processing on your behalf, central control over accounts.
Three points remain open:
- US access. OpenAI and Microsoft are US companies. The US CLOUD Act allows US authorities to compel US providers to hand over data even when it is stored outside the US. The Federal Office of Justice examined the consequences for Switzerland in a 2021 report. Opinions differ on how large the practical risk is. Professional secrecy holders need to weigh it deliberately.
- Shadow use. A company account does not stop someone from carrying on with a private account on the side.
- What goes into the prompt. Even under the best contract, the text and files entered leave your building. Whether that is acceptable for a given client depends on the engagement, not on the contract with OpenAI.
Practical options for your office
A clear written rule
The cheapest measure is a written rule everyone knows. A bare ban with no approved alternative invites people to carry on in secret.
Pseudonymisation
Names, addresses, AHV numbers and company names are replaced before input ("Client A", "property in municipality X"). This works for general questions and drafting help. It has limits: a set of facts often still points to a person, and manually redacting salary certificates or tax returns is hardly practical.
An enterprise subscription
For firms without professional secrecy in the strict sense, ChatGPT Enterprise or Microsoft 365 Copilot with European data residency can be a defensible choice. Check whether your engagement letters allow outsourcing to third parties or require you to inform clients.
Local or hybrid AI
With local AI, the language model runs on your own server. Nothing leaves the building, so the CLOUD Act and transfers abroad are not an issue for that part. In a well-built hybrid setup, documents, search index and logs stay local. Only the question and a few short excerpts go to a cloud model, and you can pseudonymise those as well or limit hybrid use to certain clients. For the hardware side, read Local AI for businesses.
Sample staff policy for AI tools
Adapt the following template to your office and have it checked by a lawyer if needed.
- Only AI tools approved by management may be used for business purposes. Currently approved: [list].
- Private accounts with ChatGPT, Gemini, Claude or similar services must not be used for client or personnel data.
- No names, addresses, AHV numbers, account numbers, salary or health data of clients or staff are entered into unapproved or public AI services.
- Documents from client files (tax returns, contracts, correspondence, salary certificates) are only uploaded to approved tools intended for that purpose.
- Approved tools may be used freely for general questions with no client reference (help with wording, explaining a legal provision).
- AI answers are drafts. Whoever uses them checks content and sources and remains professionally responsible.
- Deadlines, amounts and legal references are always checked against the original source.
- Uncertainties and possible data breaches are reported immediately to [responsible person].
- This policy is reviewed every year.
Next step
Ask your team openly which AI tools they use today, without threatening consequences. With that list you can adapt the policy above and decide which solution to approve. If a local or hybrid setup is on the table, we are happy to discuss your situation in an intro call.