CipherTax — open-source tool that redacts personally identifiable information from tax documents before sending them to AI

CipherTax detects and replaces sensitive tax data (SSN, EIN, names, email, phone, addresses, bank account numbers) with tokens while preserving financial amounts and state abbreviations that AI needs for tax calculations. The tool runs locally using Microsoft Presidio and custom tax recognizers, with PII mappings stored in an AES-128-CBC encrypted vault.

The system prevents PII leakage through five layers: PII detection, smart tokenization, encrypted local storage, a pre-send safety check that blocks API calls if any redactable PII is detected, and a test suite of 159 tests including dedicated PII leak prevention tests. Tokens use random session prefixes (e.g., [CT_a3f9_SSN_1]) to prevent collision with literal text.

CipherTax classifies tax data into five sensitivity levels (PUBLIC, INTERNAL, CONFIDENTIAL, RESTRICTED, CRITICAL) to determine redaction decisions. Data is available via pip install ciphertax on PyPI.