What is PII Pseudonymization?
PII stands for "Personally Identifiable Information". Personal data like names, email addresses, phone numbers, or AHV (Swiss social security) numbers. Pseudonymization means replacing this data with placeholders while preserving context, so no real personal data is exposed.
The difference from anonymization: With pseudonymization, the process can be reversed. Original data is preserved and reinserted after AI processing. With anonymization, the data would be irreversibly lost.
Why Pseudonymization, Not Anonymization?
For AI use cases, pseudonymization is the ideal solution. You want the AI response to come back with real data. When you ask "Write an email to Max Müller," the response should contain "Max Müller". Not "[PERSON_1]."
At the same time, the AI model should never learn that it's dealing with Max Müller. It works only with the token "[PERSON_1]" and still delivers a perfect response.
How Does It Work in Gardeo?
The process runs in four steps:
-
Detection (NER): Our PII engine uses Named Entity Recognition to identify personal data in your text. The engine recognizes categories including: personal names, company names, email addresses, phone numbers, AHV numbers, postal addresses, and IBAN numbers.
-
Consistent Replacement: Each detected entity is replaced with a consistent token. This means "Max Müller" is always replaced by the same token "[PERSON_1]". Even if the name appears multiple times. This keeps the context understandable for the AI.
-
AI Processing: The pseudonymized prompt is sent to the selected AI model. The model sees only tokens, never real personal data. Processing quality doesn't suffer. The AI understands context even with placeholders.
-
Re-substitution: The AI response is automatically re-substituted with original data. "[PERSON_1]" becomes "Max Müller" again. The result looks exactly as if you'd communicated directly with the AI. Except the real personal data never left the European legal area.
What Data Is Detected?
Gardeo automatically detects the following categories:
- Personal names: First and last names, even in different spellings
- Company names: Business names and brands
- Email addresses: All common formats
- Phone numbers: Swiss and international formats
- Social security numbers: The Swiss AHV format (756.XXXX.XXXX.XX)
- Postal addresses: Street, ZIP code, city
- IBAN numbers: Bank account numbers
- Custom entities: On the Business plan, you can define custom detection rules (e.g., case numbers, policy numbers, internal IDs)
Transparency: What Was Pseudonymized?
A unique Gardeo feature is the transparency view. After every conversation, you can see at a glance:
- Which entities were detected
- What tokens replaced them
- Whether there were false-positive detections (which you can report)
- Whether entities were missed (which you can also report)
This transparency is important not only for you as a user but also for compliance officers. They can prove at any time that personal data was properly protected.
Performance: How Fast Is Pseudonymization?
The PII engine works in real-time. Pseudonymizing a typical prompt takes under 100 milliseconds. You won't notice any difference in response time. The engine is optimized for high throughput and scales automatically with load.
Conclusion
PII pseudonymization is the heart of Gardeo. It enables you to use AI at full power without taking data protection risks. And the best part: You don't need to configure anything. It works automatically from the first message.