AI and Data Protection for Financial Firms: revFADP Checklist

Any Swiss financial services firm that uses AI will almost inevitably process personal data, and that is governed by the revised Swiss Federal Act on Data Protection (revFADP, in force since 1 September 2023). 85 percent of independent wealth managers name data protection as the biggest hurdle to adopting AI (AI Use Cases and Crypto Assets in External Asset Management, HSLU and Finup, July 2026). That large concern becomes a solvable task once you break it into concrete points. This checklist covers the ten most important duties, each in three parts: what the law says, what that means for AI, and what to do.

Why does data protection feel so big when it comes to AI?

Because three uncertainties compound: most people know the law only second hand, provider terms change constantly, and the industry thinks about liability first. The reality is simpler than it feels. The law is technology neutral, and the Federal Data Protection and Information Commissioner (FDPIC) states explicitly that the Data Protection Act applies directly to AI-based data processing (FDPIC: AI and data protection). There is no separate AI privacy law, only familiar duties applied to a new tool. For a small firm, ten points cover the essentials.

The revFADP checklist: ten points for your AI use

Work through the list in order and tick off what is already in place. For most small financial firms this is a matter of weeks, not months.

• 1. Define data classes. What the law says: it distinguishes ordinary personal data from sensitive data, such as health information in client files (Art. 5 revFADP). For AI: the class decides which environment a piece of content may enter at all. To do: introduce a simple traffic light system: public, anonymized, identifiable. The article ChatGPT in wealth management: what is allowed? shows how this works in practice.
• 2. Govern commissioned processing by contract. What the law says: if a service provider processes personal data on your behalf, this must be regulated by contract (Art. 9 revFADP). For AI: almost any provider that processes your documents or prompts is such a processor. To do: check whether your contract contains a data processing clause. Free accounts typically do not, which rules them out for client data.
• 3. Check cross-border disclosure. What the law says: personal data may in principle only be disclosed to countries that offer an adequate level of protection; the Federal Council maintains a list (Art. 16 and 17 revFADP). For AI: what matters is where the servers are and from where the data can be accessed. To do: ask every provider about data location and access from third countries. What to look for when choosing a location is covered in AI hosting in Switzerland.
• 4. Meet the duty to inform. What the law says: anyone collecting personal data informs the data subjects about the controller, the purpose and the recipients (Art. 19 ff. revFADP). For AI: if AI evaluates client documents or drafts client emails, that belongs in your privacy notice or client documentation. To do: add a short paragraph describing what you use AI for and where the data sits.
• 5. Assess whether a data protection impact assessment is needed. What the law says: processing with a high risk to personality rights can require an impact assessment (Art. 22 revFADP). For AI: AI on sensitive client data at scale can present such a risk; a limited pilot with vetted data usually stays below it. To do: briefly document why your project carries no high risk, or have an assessment reviewed professionally.
• 6. Keep a record of processing activities. What the law says: controllers keep a record of their processing activities (Art. 12 revFADP). Smaller companies are exempt under conditions; the exemption does not apply to high-risk processing and does not lift the other duties. For AI: every AI tool belongs in this record with its purpose, data types and owner. To do: a plain table is enough, and it doubles as the AI inventory the FINMA expects, see FINMA's supervisory notice 08/2024 explained.
• 7. Ensure data security. What the law says: personal data must be protected by appropriate technical and organizational measures (Data Protection Ordinance, Fedlex). For AI: role-based access, protected business accounts instead of private logins, and access logging. To do: define a minimum standard for every AI tool: multi-factor authentication, no shared private accounts, clear permissions.
• 8. Regulate retention and deletion. What the law says: personal data is kept only as long as the purpose requires; statutory retention duties, for example under FinSA, take precedence. For AI: chat histories, uploaded documents and intermediate results pile up quickly. To do: set deletion deadlines and check whether your provider deletes data completely on request.
• 9. Flag automated individual decisions. What the law says: decisions taken exclusively by automated means that produce legal effects for a person carry their own duty to inform (Art. 21 revFADP). For AI: as long as a human reviews and approves every client-relevant output, there is usually no automated individual decision. To do: put the assisted mode in writing: the AI prepares, the human decides.
• 10. Define a reporting path for breaches. What the law says: data security breaches that carry a high risk for the data subjects must be reported to the FDPIC as quickly as possible (Art. 24 revFADP). For AI: a misaddressed AI draft or a data leak at a provider can qualify. To do: decide who in the firm determines whether a report is needed and how an incident is raised internally.

From my practice: Implementation is rarely the problem; the vagueness before it is. Once the ten points are on one sheet and each point has an owner, the diffuse concern becomes a list of checkboxes. In most firms, six of the ten points are done within a month. The remaining four require management decisions, not technology.

What belongs in the contract with your AI provider?

A large part of the checklist ends up in a contract sooner or later. Six clauses should be in writing with every AI provider:

• Commissioned processing: the provider processes personal data exclusively on your behalf and under your instructions.
• Data location: named data centers and a rule for subcontractors.
• No training on your data: your inputs and documents are not used to train models.
• Deletion: defined deadlines and confirmed deletion when the contract ends.
• Incident support: the provider reports security incidents and supports you with any required notifications.
• Exit clause: you get your data back in a usable format.

Provider terms change regularly. Check the current state before signing and at least once a year afterwards.

Note: We are not a law firm. All statements on legal and regulatory topics, in particular on the revFADP and the Data Protection Ordinance, are subject to change and do not replace legal advice. For specific questions, please consult a lawyer.

Conclusion: your next step

Print these ten points and schedule a one-hour meeting with the responsible people. Go through the list, tick off what is in place, and assign an owner and a date to every open point. That turns data protection from a hurdle into a project with an end. To additionally secure day-to-day usage with a short policy, download our AI policy template for financial firms. For the full picture on use cases, costs and approach, read the 2026 guide: AI for financial services in Switzerland. And if you want to see what a protected environment for client data looks like in practice: on AI for wealth managers we show a concrete example.

FAQ

Does the revFADP also apply to a three-person office?

Yes, in principle it applies regardless of size. There is relief on the record of processing activities, but the remaining duties apply to small teams too. The advantage of small firms: ten points are decided quickly around one table.

What are the consequences of violations?

The law provides for fines against the responsible private individuals, up to CHF 250,000 in serious cases (Art. 60 ff. revFADP). In a trust-based business, reputational damage usually outweighs the fine. Assessing your specific risk belongs with a professional.

Do I have to ask clients for permission before AI processes their data?

Not in every case. Often, informing them about the processing is enough, for example in the privacy notice. Consent can become necessary when sensitive data is involved or the processing goes beyond what clients may reasonably expect. The individual case belongs with a professional.

Is this checklist sufficient as evidence for the supervisor?

No. The checklist is a working aid, not legal advice and not official guidance. What the FINMA additionally expects in terms of governance and documentation is covered in FINMA's supervisory notice 08/2024 explained.

Where do I find the binding texts?

The act is published on Fedlex (revised Data Protection Act), as is the ordinance (Data Protection Ordinance), and the FDPIC publishes guidance on an ongoing basis, including its guide to the new Data Protection Act.