AI-powered document redaction with five protection categories

Intelligent entity detection and classification

Shield uses a fine-tuned NER + LLM pipeline to detect and classify personal data in documents. Every entity is mapped to one of five protection categories: PII Standard, Sensitive PII, Client ID, Own Secrets, or Third-Party Secrets. Each category has a specific legal basis under Swiss and international law.

• PII Standard (DSG/FADP): names, addresses, dates of birth
• Sensitive PII (DSG Art. 5 lit. c): health, religion, biometrics
• Client ID (Art. 47 BankG): banking client identifiers, IBAN
• Own Secrets (Art. 162 StGB): financials, IP, strategy
• Third-Party Secrets (Contractual / NDA; Art. 162/273 StGB; UWG): NDA-covered data

Structured placeholders or full redaction

Two output modes. Structured placeholders ([PERSON_1], [COMPANY_2]) preserve document readability and entity relationships for internal review. Full redaction replaces entities with solid blocks for external publication and regulatory submissions. Both modes preserve document formatting.

Context-aware redaction

Not all personal data should be redacted equally. Shield understands context, distinguishing between a judge presiding over a case and a party named in the dispute. Smart preservation rules ensure the redacted document remains useful and legally meaningful.

Zero persistence by architecture

Shield's zero-persistence architecture means there is no data to breach. Documents are processed entirely in memory. No database stores document content. No logs capture personal data. When processing completes, the data ceases to exist.