Pr$εε$mpt introduces a formal privacy protection system for sensitive information in LLM prompts while maintaining response quality.

• Implements prompt sanitizers that transform input prompts to protect sensitive tokens
• Utilizes format-preserving encryption and differential privacy techniques
• Provides a cryptographically-inspired solution to critical security concerns in LLM APIs
• Addresses the growing challenge of privacy leakage during LLM inference

This research is critical for organizations using third-party LLM services, as it offers a practical framework for protecting proprietary or sensitive information while still leveraging powerful AI capabilities.

Pr$εε$mpt: Sanitizing Sensitive Prompts for LLMs