Ame framinPII redaction and tamper-evident audit logs for LiteLLM, no sidecars (github.com)

A new tool called Cloakwall has been introduced for LiteLLM, offering in-process PII/PHI redaction without the need for sidecar containers or external endpoints. This guardrail framework provides a simplified solution for compliance-grade audit logging and redaction of sensitive information such as email addresses, credit card numbers, and medical record numbers. Unlike traditional methods that require multiple deployments and maintenance (e.g., Presidio), Cloakwall operates as a single, dependency-free plugin that maintains privacy without exposing data outside the network. This development is significant for the AI/ML community as it streamlines the integration of security measures into machine learning models, particularly for sensitive applications in healthcare and finance. Cloakwall’s technical framework supports multiple redaction modes—masking, hashing, and partial redaction—while ensuring a tamper-evident audit log that records the types of PII detected without storing the actual values. Additionally, its low-latency operations and local data processing align well with the growing need for privacy-conscious AI solutions that can function effectively in isolated environments. Although currently not in production, Cloakwall presents a promising alternative for organizations prioritizing data security and compliance.