ReVeil is an AI-powered privacy red-team tool that tests whether removing direct PII from a document is actually enough to protect someone’s identity.

It first detects and removes direct identifiers, then uses three attacker personas — Casual Reader, Informed Investigator, and Targeted Attacker — to analyze remaining contextual clues that could enable re-identification.

ReVeil validates AI-generated claims using quote-or-drop evidence validation, calculates an estimated heuristic privacy risk, identifies attack surfaces and evidence chains, suggests contextual hardening, and attacks the hardened document again to measure privacy improvement.

The system runs with FastAPI and a local Ollama/Qwen3 1.7B AI model, with a 7-screen interface covering analysis, attack, risk reporting, hardening, and before/after regression testing.