The nDSG and the AI Challenge
Since September 1, 2023, the new Swiss Data Protection Act (nDSG) has been in effect. It introduces stricter rules for handling personal data. And poses a critical question for companies using AI tools like ChatGPT or Gemini: How can we use AI productively without violating data protection regulations?
The answer isn't straightforward. Every prompt that employees enter into an AI tool may contain personal data: customer names, contract numbers, email addresses, internal project names, or even social security numbers.
The Problem with Conventional AI Tools
When an employee pastes a customer email into ChatGPT or summarizes a medical letter using a cloud-based AI tool, that data leaves Switzerland. It's transmitted to servers in the US or EU. Often without the affected individuals knowing.
This isn't just a technical problem. It's a legal one. The nDSG requires a legal basis for transferring personal data abroad. In many cases, this basis is missing. Especially when AI tools are used informally without IT approval (so-called "Shadow AI").
According to the Deloitte Switzerland study "The rapid arrival of generative AI in Switzerland" (2023, n=1,002), 61% of Swiss computer-based workers already use generative AI in their daily work. But 61% also report that their company has no clear policies for it. This creates enormous compliance risk.
The Risks in Detail
- Data protection violations: Personal data could flow abroad without legal basis.
- Reputation damage: A data leak with customer data can permanently damage trust.
- Fines: The nDSG provides for fines of up to CHF 250,000 for responsible individuals.
- Contractual risks: Many customer contracts and NDAs prohibit sharing data with third parties.
The Solution: Automatic Pseudonymization
Gardeo solves this with a unique approach: automatic PII pseudonymization. Before your prompt reaches the AI model, all personal data is automatically detected and replaced with tokens.
Here's how it works step by step:
- You write your prompt as usual. With real names, emails, and data.
- The PII engine detects all personal data in real-time.
- Detected data is replaced with consistent pseudonyms (e.g., "Max Müller" → "[PERSON_1]").
- The pseudonymized prompt is sent to the AI model.
- The response is re-substituted with the original data.
The result: You get full AI performance. But real personal data never leaves the European area. Processing happens in German data centers (Frankfurt and Brandenburg), fully nDSG- and GDPR-compliant.
Multi-LLM: The Best AI for Every Task
Another advantage: Gardeo provides access to leading AI models (OpenAI GPT, Anthropic Claude, Google Gemini, Mistral, and more) through a single interface. You can switch models per conversation and aren't locked into one provider.
This also means: If a provider changes their data privacy policies, you can immediately switch to another model. Without interrupting your workflows.
Conclusion
Privacy-first AI is not a compromise. It's a prerequisite for responsible AI use in Switzerland. Companies investing in privacy-compliant AI solutions today create a sustainable competitive advantage: They can roll out AI faster, onboard employees without concerns, and demonstrably act responsibly toward customers.
Gardeo makes this possible. With EU hosting in Germany, automatic pseudonymization, and multi-LLM flexibility — built in Switzerland.