HomeLearnRAG under the Swiss nFADP (nDSG)
Learn / Compliance

RAG under the Swiss nFADP (nDSG)

How to run retrieval-augmented generation in compliance with the Swiss Federal Act on Data Protection (nFADP/nDSG): legal basis, cross-border transfers, technical measures and a practical checklist.

Last updated: 2026-07-28

The revised Swiss Federal Act on Data Protection (nFADP; German nDSG, French nLPD) has applied since 1 September 2023 and governs any processing of personal data by private companies and federal bodies. A RAG system that ingests HR files, client dossiers, claims records or correspondence is processing personal data — and the classic failure mode is shipping that data to a foreign cloud AI without a lawful transfer basis.

Where RAG touches the nFADP

  • Processing principles (Art. 6). Proportionality and purpose limitation apply to what you ingest: index what users need to query, not everything you can.
  • Cross-border disclosure (Art. 16–17). Sending documents or embeddings to a provider abroad is a disclosure. It requires an adequate jurisdiction or safeguards — and US CLOUD Act exposure complicates both. Keeping the whole pipeline in Switzerland removes the question.
  • Data security (Art. 8). Encryption at rest and in transit, access control and logging must match the sensitivity of the corpus — for special categories (health, legal files), that bar is high.
  • Access & deletion rights (Art. 25 ff.). You must be able to find and delete a person's data — including its chunks, vectors and graph entries.

Technical measures that satisfy the act

nFADP requirementTechnical answer in SovraRAG
Data security (Art. 8)AES-256-GCM envelope encryption per tenant, applied before storage; TLS in transit; operator-blind design
No unlawful cross-border transferInference, vectors, archive and backups exclusively in Swiss datacenters
Accountability & auditabilityComplete query and audit logs, configurable retention
Deletion rightsDocument-level deletion cascades to chunks, vectors and graph entries
Access limitationPer-document ACLs enforced at the retrieval layer

Practical checklist before you deploy RAG

  • Classify the corpus: does it contain personal or sensitive personal data?
  • Map every subprocessor in the pipeline — including the embedding and reranking APIs, which are easy to overlook.
  • Verify where inference runs and under which jurisdiction the operator is incorporated.
  • Demand per-tenant encryption and ask who can technically decrypt.
  • Test the deletion path end-to-end before go-live.

This article is general information, not legal advice.

See it on your own documents

Get a guided demo and a proof-of-concept with your corpus — on Swiss infrastructure from day one.