𝗚𝗗𝗣𝗥 𝘃𝘀 𝗙𝗔𝗗𝗣: 𝗪𝗵𝘆 𝗦𝘄𝗶𝘀𝘀 𝗔𝗜 𝗽𝗿𝗼𝗷𝗲𝗰𝘁𝘀 𝗯𝗹𝗲𝘄 𝘁𝗵𝗲𝗶𝗿 𝗯𝘂𝗱𝗴𝗲𝘁𝘀
A Zurich chatbot startup paid a CHF 150,000 fine in May 2026.
They reused training data without a specific FADP impact assessment. They thought GDPR rules were enough. They were wrong.
The regulator found that the company trained models nightly on raw customer chats. 3% of these records contained high-risk personal data. The company never masked this data.
The mistake cost more than just a fine. The company had to shut down its service for 48 hours. This caused customer churn and a loss of €12,000 in revenue.
The problem lived in the code. The "redact-PII" step only worked for GDPR. It did not recognize the new FADP requirements.
Swiss law changed in 2026. Article 29 now requires a Data Impact Simulation before you deploy any model.
GDPR allows you to perform assessments after a model is live. FADP requires you to do it before training begins.
Here is how the requirements differ:
• Data Inventory: FADP requires mandatory tagging for AI risk. • DPIA: FADP requires sign-off before training starts. • Model Training: FADP requires an AI Impact Simulation. • Redaction: FADP makes automatic redaction of high-risk fields mandatory. • Deployment: FADP requires a signed impact PDF and policy-as-code gates.
The AI Impact Simulation adds five days to your release cycle. A survey of 78 Swiss AI projects shows this adds 3.2 weeks per release. This increases project costs by 27%.
Most MLOps tools only support GDPR. They lack the simulation engine needed for FADP. This forces teams to use manual spreadsheets.
To avoid these costs, you must integrate compliance into your CI/CD pipeline.
Do these three things:
- Treat the AI Impact Simulation as your first gate.
- Build lineage tagging into your data ingestion layer.
- Make redaction mandatory for all scheduled jobs.
One firm used a pre-built FADP template. They reduced their time-to-market by 22%. They moved from paperwork back to feature work.
Stop treating compliance as an afterthought. Build it into your code.
Optional learning community: https://t.me/GyaanSetuAi