Here's what it looks like in practice:

1. Select a frontier model
2. PrivDF warns you and offers to sanitize
3. Review detected PII (names, emails, SSNs, etc.)
4. AI sees "[NAME]" and "[EMAIL]" instead of real data

Here's Gemini 3 Pro analyzing the same document.

Notice the citation: "CONSULTANT: [NAME]..."

The AI is working with sanitized text. Your actual client names, SSNs, addresses? Never sent.

Full analysis power. Zero privacy risk.

This is what the AI actually sees:

"Agreement between [NAME] and [NAME], SSN: [SSN], contact: [EMAIL]..."

All your sensitive data replaced with type labels. The AI can still analyze structure, extract terms, answer questions.

It just can't see the actual values.

Our multi-model scanner detects:

📧 Emails & phone numbers
🔢 SSNs & credit cards
📍 Physical addresses
👤 Names (with Dr., Mr., etc.)
📅 Dates of birth
🔑 Account numbers

You review everything before redaction - nothing is automatic.

Step 1: PrivDF detects you're about to use a frontier model (Gemini, Claude, GPT)

It prompts you to sanitize first - scanning with PRIVATE AI models where your data never leaves Venice's secure servers.

You choose what gets redacted before anything is sent.