Every detected entity gets a unique typed placeholder — [NAME_1], [EMAIL_2], [PHONE_1] — preserving AI context while eliminating PII.
The regex engine identifies entities in your text.
Each entity is replaced: "John Doe" → [NAME_1], "john@co.com" → [EMAIL_1].
Send tokenized text to any AI. Context preserved, privacy protected.
![Zero-Trust PII Detection pipeline — raw document with PII passes through the PrivacyScrubber shield, producing anonymized tokens like [NAME_1] and [EMAIL_2]](/features/pii-detection-hero.png)
PrivacyScrubber's typed token system processes your data 100% locally in browser memory. No server ever sees your content — verified by our Airplane Mode test. This Zero-Trust Data Sanitization (ZTDS) architecture meets enterprise security standards out of the box.
When an NDA mentions 5 different parties, typed tokens keep each identity distinct: [NAME_1] is the buyer, [NAME_2] is the seller. AI can reason about relationships without knowing real names.
Sales teams paste CRM notes into AI for follow-up emails. Typed tokens let the AI reference "Dear [NAME_1]" and discuss "[COMPANY_1]" without exposing real client data.
Support teams use AI to draft responses. Tokens like [EMAIL_1] and [NAME_1] let AI compose personalized replies that get un-masked before sending.
Researchers anonymize interview transcripts for AI analysis. Each participant becomes [NAME_N], maintaining narrative coherence while protecting identities.
Tokens follow the format [TYPE_N] where TYPE is the entity category (NAME, EMAIL, PHONE, ID, CUSTOM) and N is a sequential number. For example: [NAME_1], [EMAIL_2], [PHONE_1].
Generic redaction (e.g., replacing everything with [REDACTED]) destroys context. Typed tokens preserve entity relationships so AI can reason about "the email sent from [EMAIL_1] to [NAME_2]."
Yes. The same person is always [NAME_1] throughout your session. This consistency lets AI maintain narrative coherence across multiple text blocks.