PII Redaction for LLM Prompts
PII redaction for LLM prompts is a process that ensures sensitive information is protected before queries are sent to language models. OpenSmartRoute provides tools to perform prompt sanitisation and redaction, enhancing security and compliance.
Overview of PII Redaction in OpenSmartRoute
OpenSmartRoute incorporates PII redaction as part of its input handling process. The InputGuard component can be configured to detect and redact personally identifiable information (PII) from prompts before they reach the language model. This process helps prevent leakage of sensitive data and maintains user privacy.
How PII Redaction Works
The redaction process involves several steps:
- Detection: The system scans the prompt for PII using heuristics and learned detectors.
- Redaction: Identified PII is replaced or obscured to prevent exposure.
- Sanitisation: The prompt is cleaned to remove any confounder gadgets or injection risks.
This process is configurable and can be tailored to specific security requirements.
Configuring PII Redaction
Here is an example of how to enable PII redaction in your routing setup:


