Can a Swiss law firm use ChatGPT or Copilot with client data?

Yes, if you use a business AI service under contract that does not train on your data, configured so the provider is bound to confidentiality and only the right people see the right files. No, if anyone in the firm is pasting client information into a personal ChatGPT account.

I run IT and security operations at a FINMA-regulated bank, and this is the same question regulated firms of every size are working through. The good news for a small law firm or Treuhand is that the answer does not need new law or a large budget. It needs the right plan, a few settings and a short written rule. This is not legal advice; for questions about your own professional obligations, speak to your counsel or your bar association.

The real risk is the personal account

Consumer ChatGPT accounts (Free, Plus and Pro) can use your conversations to train OpenAI's models unless the user switches this off in the data controls. OpenAI's business plans work the other way round: ChatGPT Business, Enterprise and the API are not used for training by default, and OpenAI offers a data processing addendum for them. [3, 4]

Microsoft takes a similar line for Microsoft 365 Copilot: prompts, responses and the data Copilot reaches through Microsoft Graph are not used to train the foundation models, and Copilot falls under Microsoft's existing commitments for Microsoft 365 business customers, including the EU Data Boundary. [5]

So the first control is simple: client data only goes into a business account the firm controls, never a personal one. Before anything else, find out whether anyone is already using a personal account for work, usually with the best of intentions. That is the gap to close first.

What Swiss data protection law says about AI

The Federal Data Protection and Information Commissioner (FDPIC) has been clear that the revised Federal Act on Data Protection, in force since 1 September 2023, applies directly to processing that uses AI. You do not have to wait for an AI law. [1]

In practice, the FDPIC's position means three things for a small firm:

• Transparency. Be open about the purpose of AI-supported processing, how it works and what data it uses. A line in your privacy notice and engagement terms is a sensible start.
• People stay in charge of decisions. The law gives individuals rights around automated individual decisions, including asking for a human review. Keep a person reviewing anything AI drafts before it reaches a client or decides something about one.
• High-risk processing needs a data protection impact assessment. Drafting help and document search for a small firm is unlikely to be high risk, but if you build something that profiles people or decides outcomes, assess it first.

Professional secrecy is the higher bar

For lawyers, the stricter rule is professional secrecy under Art. 321 of the Swiss Criminal Code and the duties in the Lawyers Act (BGFA). The leading analysis of cloud use, prepared for the Swiss Bar Association and published in the Anwaltsrevue in 2019, concluded that lawyers may in principle use cloud services. The provider is treated as an auxiliary person (Hilfsperson), so giving it access is not an unauthorised disclosure, provided the lawyer meets the duties that come with it. [2]

Those duties translate well to AI services, which are cloud services by another name:

• Choose, instruct and supervise the provider carefully.
• Bind the provider to confidentiality by contract and limit its use of the data to providing the service.
• Keep the circle small for sensitive matters and add technical safeguards.
• Assess foreign providers case by case, weighing how sensitive the data is and how likely foreign access is.
• Consider client consent. The authors did not see it as strictly required, but recommended it as an extra safeguard, for example in the engagement letter.

One update to that 2019 paper: it mentions Privacy Shield, which no longer exists. Since 15 September 2024, Switzerland recognises the United States as adequate only for recipients certified under the Swiss-US Data Privacy Framework. Microsoft, among others, is certified. [6]

The setup I would put in place

1. Pick one business AI service the firm controls. If you already run Microsoft 365, Copilot keeps prompts and files inside the tenant you already trust. Otherwise, a ChatGPT business plan with the data processing addendum signed. Not a mix of personal accounts.
2. Check the contract terms against the duties above: confidentiality, purpose limitation, data location options and sub-processors. Note that OpenAI's data residency options are offered on Enterprise and API plans, not on ChatGPT Business. [4]
3. Fix permissions before switching Copilot on. Copilot only shows people what they can already open, which means an over-shared SharePoint or Teams site becomes very easy to search. Tidy access to client folders first. [5] More in rolling out Microsoft 365 Copilot in a small Swiss firm.
4. Write a one-page AI rule. Which tool, what may go in, what may not, and that a person checks every output before it leaves the firm. A template: a one-page AI policy for a small Swiss firm.
5. Tell clients. A short paragraph in your engagement terms and privacy notice covers the transparency point and the consent recommendation together.
6. Record it. Add the AI service to your record of processing activities and your list of processors.

None of this is heavy work for a small firm, and it turns AI from a quiet risk into something partners can sign off. For the workflows themselves, see AI and automation for law firms. If you want help with the setup, the AI Opportunity Audit includes the data protection setup for each workflow it recommends.