LLM PII Redaction & DLP — Pre/Post Filter
Detect and redact emails, SSNs, credit cards, secrets, and custom identifiers on both sides of every LLM call. Compliance-ready logs, per-tenant policies, and reversible tokenization for trusted workflows.
What gets redacted
- Emails, phone numbers, postal addresses
- Government IDs — SSN, passport, national ID, tax ID
- Payment data — credit cards (Luhn-validated), IBAN, SWIFT
- Secrets — API keys, JWTs, AWS/GCP/Azure credentials, private keys
- Custom regex per tenant — internal IDs, ticket numbers, account refs
Pre vs post filter
The pre-filter redacts PII before it reaches the model — protects against training data leakage and provider-side logging. The post-filter redacts PII before responses reach the user — protects against the model echoing sensitive data from retrieved context.
Tokenization
Optional reversible tokenization replaces PII with stable opaque tokens, lets the LLM reason over them, and restores the original values in the final response. Useful for support and CRM workflows.