Run AI assistants and agents under Switzerland’s revised data protection act

Switzerland has no separate AI law yet: the revised Federal Act on Data Protection (revFADP), in force since 1 September 2023, applies directly to AI. For an assistant or agent that means stating its purpose and data sources, telling people when they are talking to a machine, letting them have an automated decision reviewed by a person, and a data protection impact assessment where the risk is high. Knowlytica produces the documentation and builds the controls into the system; your legal counsel signs off. The first 30-minute call is free.

Status: available now. We produce the documentation and controls; legal sign-off stays with your counsel.

From inventory to log
Inventory
Art. 12
Transparency notice
FDPIC guidance
Automated decision check
Art. 21
Impact assessment if high risk
Art. 22–23
Cross-border check
Art. 16
Audit log and quarterly review
Ongoing
Diagram: the six revFADP steps for an AI system, from the inventory under Art. 12, the transparency notice, the automated decision check under Art. 21, the impact assessment under Art. 22–23 and the cross-border check under Art. 16, to an ongoing audit log and quarterly review.

How to make an AI assistant compliant with the revFADP

Six steps, citing the article of the revised act each one rests on. The Swiss data protection commissioner (FDPIC) confirmed in May 2025 that these rules apply to AI as they stand.

  1. Step 01

    Inventory every AI use

    For each tool or agent, record its purpose, the personal data it touches, the vendor and where the data is processed. This feeds the record of processing activities under Art. 12.

  2. Step 02

    Write the transparency notice

    The FDPIC expects the purpose, how the system works and its data sources to be disclosed. People must be told when they are communicating with a machine, and whether their input is used to improve the system.

  3. Step 03

    Find automated individual decisions

    If the system alone makes a decision with legal effects or a significant impact on a person, Art. 21 requires telling them and, on request, letting them state their view and have a person review it. We build that review step into the agent.

  4. Step 04

    Run an impact assessment where risk is high

    Art. 22 requires a data protection impact assessment when processing is likely to create a high risk, for example profiling or large volumes of sensitive data. If a high risk remains after the planned measures, Art. 23 requires consulting the FDPIC first.

  5. Step 05

    Check where the data goes

    Under Art. 16, a vendor processing outside Switzerland needs a country on the Federal Council’s adequacy list, US certification under the Swiss–US Data Privacy Framework, or contractual safeguards such as standard contractual clauses.

  6. Step 06

    Log and review

    Keep a log of what the system read, decided and changed. Review the inventory and the notice every quarter, and whenever a new tool or data source is added.

The part that makes it fail

Personal AI accounts are where compliance breaks

The usual gap is not in the system you documented but next to it: staff pasting client data into a personal AI account. Personal plans have their own training and retention settings, so the notice you published no longer matches what happens to the data.

The fix is organisational more than technical: a business workspace for everyone who needs AI, a one-page rule on which data may go where, and the inventory from step 1 kept current.

Personal accountSettings chosen by each employeeNot in your inventory, not covered by your notice.
Business workspaceSettings chosen by the organisationListed, contracted and described in the notice.

What Knowlytica adds to the FDPIC’s guidance on AI

The guidance says what the revFADP expects. These are the documents and controls that show you meet it.

An AI inventory template

Purpose, data, vendor and location per system, ready for the Art. 12 records.

Notice text in three languages

English, German and French, including the machine disclosure.

Human review built in

Agents pause for a named person’s approval before decisions that affect people.

An impact assessment draft

Risks, measures and remaining risk, written for your counsel to finalise.

A transfer table

Every vendor with its processing country and legal basis under Art. 16.

An audit log

Who did what, when and with which data, exportable on request.

Frequently asked questions

Is there a Swiss AI Act?

No. On 12 February 2025 the Federal Council decided to ratify the Council of Europe’s AI Convention and adapt existing laws sector by sector instead of passing one AI act. A consultation draft is due by the end of 2026; until then the revFADP is the main rulebook.

How much does it cost?

The first 30-minute call is free. The documentation and controls are quoted in a written proposal for the agreed scope.

What isn’t it?

It is not legal advice and not a certificate. We produce the technical documentation and build the controls; your counsel or data protection officer approves them.

Does this also cover the EU AI Act?

Partly. The inventory and the logs serve both, but the EU AI Act adds duties by role and risk class. See the EU AI Act for Swiss companies.

A customer asked why our chatbot rejected their request. What do we owe them?

If the chatbot alone decided and the decision affects them significantly, you must tell them it was automated and, on request, have a person review it (Art. 21). If nobody can say why it decided, the system was not logging its inputs; the agents we build log the inputs of every decision.

What are the penalties?

Intentional breaches of key duties can be fined up to CHF 250,000. Under the revFADP the fine is imposed on the responsible individual, not on the company.

Also works with

If you also serve EU customers, read the EU AI Act for Swiss companies. The controls described here are built into our custom MCP servers and our Claude rollouts, and into the client agents on AI agents for marketing agencies. How we handle data ourselves is on the trust page, and every service is on the Knowlytica home page.