Appearance
Before vs After — GhostAI Protection Demo
The simplest way to understand what GhostAI does: send the exact same prompt through unprotected AI and through GhostAI, side by side. The AI gives the same quality answer both ways — but only one path exposes your customer's SSN, credit card number, and home address to a third-party model. This demo is ideal for any audience, technical or non-technical.
Pre-Demo Checklist
- [ ] GhostAI instance running with demo data
- [ ] Comparison mode enabled
- [ ] No vertical pre-selected (default mode)
Demo Script
Chapter 1: Side-by-Side Comparison
Duration: 3 minutes Story: A customer service agent needs help responding to a complaint. The prompt contains a customer's name, email, credit card number, SSN, date of birth, and home address. In comparison mode, the viewer sees raw AI exposure vs GhostAI protection side by side.
Input:
Help me respond to this customer complaint:
From: Sarah Thompson (sarah.thompson@gmail.com)
Account: 4539-7821-0043-5567
SSN on file: 287-65-4321
DOB: March 15, 1987
Address: 1847 Oak Boulevard, Apt 4B, Denver CO 80202
She says she was charged $342.99 twice for order #WH-284719 and wants a refund plus compensation for the overdraft fee.Expected Output:
- Left panel (raw): All PII visible — name, email, card number, SSN, DOB, address sent to model in cleartext
- Right panel (GhostAI): All PII tokenized before reaching the model; response quality identical
- Both panels show a professional customer service response
Talking Points:
"Same prompt. Same AI model. Same quality response. But look at what went to the model on the left — her name, SSN, credit card, home address, all of it in cleartext. On the right, GhostAI tokenized every piece of sensitive data before the model ever saw it. The customer gets the same great answer. The difference is who else saw her data."
Chapter 2: What the AI Model Actually Receives
Duration: 2 minutes Story: Switch to a healthcare scenario showing a patient record with PHI. The PII detection results panel reveals exactly what the AI model received — every identifier replaced with a reversible token.
Input:
Show me the PII detection results for this text: Patient Robert Kim, MRN 9920381, was seen by Dr. Emily Watson at Mount Sinai Hospital on 02/14/2025. Diagnosis: Type 2 Diabetes (E11.9). Insurance: BlueCross policy BC-449281003. Emergency contact: Lisa Kim at (415) 555-0167.Expected Output:
- PII detection panel shows each identified entity: patient name, MRN, provider name, facility, date, diagnosis code, insurance ID, emergency contact name, phone number
- Tokenized output displayed with token → original mapping
- Confidence scores for each detection
Talking Points:
"This is exactly what the AI model received. Every name, every ID number, every phone number has been replaced with a reversible token. The model sees the structure — it knows there's a patient, a diagnosis, an insurance policy — but it never sees the real values. When the response comes back, GhostAI restores the originals for the authorized user."