Switzerland’s revised Federal Data Protection Act (DSG) has been in force since 1 September 2023. One consequence that is frequently overlooked: the DSG is formulated in technology-neutral terms, which means it applies directly to AI-based data processing - without requiring separate AI-specific legislation. The Swiss Federal Data Protection and Information Commissioner (FDPIC/EDÖB) confirmed this explicitly in an updated guidance document issued on 8 May 2025. Law firms and in-house legal teams using AI tools are therefore operating within a defined legal framework - provided the tools they use and their internal processes meet DSG requirements.
What the DSG means in practice for AI applications
The DSG protects personal data of natural persons. Since the revision, this includes genetic and biometric data as specially protected categories. For AI systems, this matters because personal data can be processed at multiple points in the AI lifecycle: as training data, for contextualisation, as user input, and as AI-generated output.
The DSG requires organisations and public bodies deploying AI to take concrete measures. The most relevant ones are:
Transparency obligation: Anyone carrying out AI-based data processing must disclose the purpose, functionality, and data sources. Users interacting with AI language models have the right to know whether they are communicating with a machine and whether their input is used for model training.
Privacy by Design and Privacy by Default: Art. 7 DSG requires data protection to be considered from the earliest development and planning stages - not only at rollout. For external AI tools, this means assessing whether a system meets these requirements before procurement.
Data Protection Impact Assessment (DPIA): Art. 22 DSG requires a DPIA when data processing carries a high risk to the personality rights or fundamental rights of those affected. High-risk AI applications are permitted in principle, but only with appropriate protective measures.
Data breach notification: Under Art. 24 DSG, data breaches must be reported to the FDPIC as quickly as possible – but only if they are likely to lead to a high risk to the personality rights or fundamental rights of those affected. Unlike the GDPR (72-hour deadline triggered by any risk), the DSG sets a higher threshold and no fixed deadline.
Prohibited AI applications under the DSG
The DSG prohibits applications specifically designed to undermine privacy and the right to informational self-determination. The FDPIC explicitly cites mass real-time facial recognition in public spaces and social scoring - meaning comprehensive behavioural monitoring and rating of individuals - as examples. These practices are primarily observed in authoritarian state systems, but the boundary is still relevant for compliance purposes.
Switzerland’s regulatory framework: DSG instead of an AI Act
Unlike the EU, Switzerland does not yet have dedicated AI legislation. The Federal Council signed the Council of Europe Convention on AI and Human Rights (adopted 17 May 2024) on 27 March 2025. Ratification is planned, and the consultation proposal is expected by the end of 2026. Lead responsibility for the legal regulatory work sits with the Federal Department of Justice and Police (EJPD) and the Federal Office of Justice (BJ), together with the DETEC and the FDFA. The Federal Office of Communications (BAKOM) prepared the initial situational analysis (February 2025) and is now working on non-binding accompanying measures.
Switzerland’s stated approach has three goals: strengthening innovation, protecting fundamental rights including economic freedom, and building public trust in AI systems. Until a Swiss-specific AI law enters into force, the DSG remains the primary legal instrument.
For organisations with EU exposure: the DSG is closely modelled on the GDPR, which contributed to the EU’s renewed adequacy decision for Switzerland (confirmed on 15 January 2024). DSG compliance aligns with many GDPR requirements in practice - but the two frameworks are not fully equivalent.
DSG vs. GDPR: Key differences for legal teams
Anyone advising clients with EU exposure or evaluating AI tools across a corporate group inevitably compares the two regimes. The frameworks are closely related but diverge at several decisive points:
Topic | Swiss DSG | EU GDPR |
|---|---|---|
Fines | Up to CHF 250,000 against responsible natural persons (e.g. management) | Up to EUR 20 million or 4% of global annual turnover against the company |
Breach notification deadline | “As quickly as possible”, no fixed deadline | 72 hours from awareness |
Notification threshold | Only when high risk to personality / fundamental rights | Already at any risk |
Underlying principle | Permission with reservation of prohibition | Prohibition with reservation of permission |
Data protection officer | Optional | Mandatory in many cases (Art. 37 GDPR) |
Scope of protection | Natural persons only | Natural persons only |
Right of access | Art. 25 DSG | Art. 15 GDPR |
For AI deployments this means: organisations with GDPR-compliant processes typically meet most DSG obligations – but they still need to understand Swiss specifics around personal liability of management, notification deadlines, and the underlying burden of proof.
Why legal teams need to pay particular attention
Legal teams handle highly sensitive data every day: contract documents, client information, due diligence files, HR records. When this content is fed into AI tools, data protection questions arise immediately: where is the data stored? Who has access? Are inputs used for training?
The AXA KMU-Arbeitsmarktstudie 2025 puts numbers on the gap: 34% of Swiss SMEs deliberately integrate AI, a further share experiments with or has occasional experience using AI, and 29% have never used it. Among the SMEs that do use AI, however, only about one in three has defined clear internal data protection rules for AI use. In most organisations, employees independently decide which tools to use and what data to enter. From a data protection standpoint, that is a significant risk.
How the FDPIC views data protection in AI training is illustrated by its preliminary investigation into X (formerly Twitter) and the AI model Grok: X had introduced an opt-out for using public posts in AI training on 16 July 2024. On 20 March 2025, the FDPIC closed the investigation, concluding that X meets the requirements of the DSG with this opt-out option. The takeaway: the FDPIC accepts an effective opt-out as a permissible basis for using publicly accessible data for AI training – but expects transparency about the practice.
What DSG-compliant AI tools for legal teams look like
An AI tool used in a legal context should concretely meet the following criteria:
No data retention: Inputs and documents are not stored permanently. CASUS, a Swiss legal AI platform for law firms and in-house legal teams, operates with zero data retention - no data is stored after processing.
No human review: Content is not viewed by human staff at the provider. CASUS offers an abuse monitoring opt-out, meaning no human review of user inputs takes place.
Hosting in Switzerland or the EU: Data must not be transferred to third countries without an adequate level of data protection. CASUS hosts exclusively in Switzerland and the EU - no data transfer to the US.
Transparency about processing: Users should be able to understand what happens to their inputs.
These are not marketing promises - they are technical and contractual prerequisites for legally compliant AI use in a legal setting.
Practical implications for law firms and legal teams
Any firm using AI tools for contract analysis, due diligence, or legal research should address three internal questions:
First: what data is being entered into the tool? If it includes personal data of natural persons - which is almost always the case in legal documents - the DSG applies.
Second: is a DPIA required? For high-risk processing, yes. For standard contract analysis without profiling, the risk level is often lower, but a review is still recommended.
Third: does the organisation have internal AI usage policies? According to the AXA study, roughly two in three SMEs that already use AI lack clear data protection rules for AI use. DSG violations can result in fines of up to CHF 250,000 against responsible natural persons (typically management) for wilful breaches – on top of reputational damage.
CASUS’s AI Data Room supports the processing of large document volumes and can detect personal data such as names, email addresses, IDs, and bank details - prioritising sensitive data categories. This supports preparation for anonymisation before documents are shared further.
For legal research on data protection questions, CASUS offers a Legal Research mode that draws on statutes, case law, and legally reliable sources, delivering structured, traceable outputs rather than generic internet answers.
CASUS as a data-protection-compliant option
CASUS is a Swiss legal AI platform that works directly in Microsoft Word or as a web app. The platform includes modules for contract analysis (Risk & Quality Review), document comparison (Benchmark), legal proofreading (Proofread), and AI-powered chat with documents (AI Chat).
The architecture is explicitly designed for the Swiss and European legal market: hosting in Switzerland and the EU, no data transfer to the US, zero data retention, and no human review. This means the platform meets the core technical requirements for DSG-compliant use in a legal context.
Firms looking for an AI solution that fits within Swiss data protection law can try CASUS for free at app.getcasus.com/signup.
FAQ
Does Switzerland’s DSG apply to AI applications?
Yes. The DSG is formulated in technology-neutral terms and applies directly to any AI-based data processing. The FDPIC confirmed this explicitly in updated guidance issued on 8 May 2025.
Does Switzerland need a dedicated AI law?
Not yet in force. Switzerland signed the Council of Europe Convention on AI on 27 March 2025. The consultation proposal is expected by the end of 2026. Until then, the DSG remains the governing instrument.
When is a Data Protection Impact Assessment (DPIA) required for AI projects?
A DPIA under Art. 22 DSG is required when data processing carries a high risk to the personality rights or fundamental rights of those affected. This applies particularly to AI applications involving extensive profiling or automated individual decision-making.
Which AI applications are prohibited under the DSG?
Applications specifically designed to undermine privacy and informational self-determination are prohibited. The FDPIC cites mass real-time facial recognition and social scoring as clear examples.
Can a law firm enter client data into AI tools?
It depends on the tool. The key factors are: where is the data stored, are inputs used for model training, and is there human review? A law firm must ensure that any tool it uses is DSG-compliant - in particular, no data retention and no US hosting.
What does “zero data retention” mean in an AI context?
Zero data retention means that input data and documents are not stored permanently after processing. This is a technical requirement for meeting the DSG obligations around the right to erasure and purpose limitation.
How does Switzerland’s DSG differ from the EU GDPR?
The DSG is closely modelled on the GDPR, which led to the renewed EU adequacy decision for Switzerland in 2024. Practically relevant differences: under the DSG, fines of up to CHF 250,000 target responsible natural persons (typically management), while the GDPR fines companies up to EUR 20 million or 4% of global annual turnover. The DSG requires breach notification “as quickly as possible” with no fixed deadline and only at high risk, whereas the GDPR mandates 72 hours triggered by any risk. The DSG follows a principle of permission with reservation of prohibition; the GDPR follows the inverse. A data protection officer is optional under the DSG, but mandatory in many cases under the GDPR.
How can legal teams reduce AI-related data protection risks internally?
Practical steps include: assessing every AI tool for DSG compliance before use, establishing clear internal policies on permitted data inputs, training staff on applicable rules, and - where possible - anonymising personal data before AI processing.
