Risk Mirror – Deterministic AI Security for Sensitive Data

Summary: Risk Mirror is a developer toolkit that intercepts user input before it reaches LLMs like OpenAI or Anthropic, removing sensitive data using deterministic algorithms. It detects 152 types of personally identifiable information (PII) and high-entropy strings without relying on AI inference, ensuring consistent and reliable data redaction.

What it does

Risk Mirror sits between users and LLMs to strip sensitive data using custom classifiers, entropy detection, and deterministic rules without AI models. It supports data types from US Social Security numbers to HIPAA identifiers.

Who it's for

Developers integrating LLMs who need to prevent sensitive data leakage in user inputs.

Why it matters

It eliminates the uncertainty of AI-based data filtering by providing 100% consistent, non-ML redaction to protect sensitive information before processing.