AI PII/PHI Redaction Agent

How Cassidy automates PII/PHI Redaction using AI

Step 1: Ingest and normalize claim files

The Workflow triggers when documents land in a watch folder or claim system. Cassidy pulls the entire claim folder, dedupes files, applies OCR to scanned PDFs and images, and preserves pagination for page-level traceability.

Step 2: Classify documents automatically

Cassidy identifies document types—medical records, FNOL forms, adjuster notes, EOBs, police reports, UB-04/CMS-1500 forms—to apply the right detection rules for each.

Step 3: Detect sensitive data with hybrid rules and NLP

Configurable rule packs combine regex patterns for structured identifiers (SSN, DOB, driver's license, VIN) with NLP-based entity recognition for names, clinical terms, and context-sensitive content. Confidence scoring flags low-certainty hits for review.

Step 4: Apply jurisdiction and line-of-business rule packs

Cassidy applies HIPAA Safe Harbor, CPRA, GDPR, DPPA, and state-specific requirements based on the production type—whether it's a Workers Comp subpoena, vendor share, or reinsurance review.

Step 5: Burn-in redactions and flatten outputs

Irreversible masking is written directly into the output rendition. Hidden content, annotations, and alternate layers are removed to prevent accidental disclosure.

Step 6: Validate with Human-in-the-Loop review

Reviewers use Q&A to verify redactions—"Show all pages with SSNs," "Confirm all DOBs are masked"—and approve or adjust before final production.

Step 7: Export with audit trail and deliver

Cassidy generates paginated redacted PDFs with index sheets and page-level audit logs documenting every detection, rule, confidence score, and approver. Outputs integrate directly with Guidewire, Duck Creek, OnBase, or SharePoint for delivery.