OneShield Privacy Guard - Deployable Privacy Solutions for LLMs

Learn about OneShield Privacy Guard, a comprehensive framework designed to address privacy challenges in Large Language Model deployments through this 14-minute conference talk from PEPR '25. Discover how this solution implements context-aware guardrails across the entire LLM lifecycle, including input preprocessing, inference, and output sanitization, while utilizing automated risk assessment for continuous improvement. Explore two practical deployments: an enterprise-scale multilingual system that enhanced PII detection accuracy and optimized privacy-utility tradeoffs in high-volume environments, and an open-source implementation that reduced manual intervention in privacy-sensitive pull requests by 30% while maintaining compliance precision. Gain insights into the technical architecture, deployment challenges, and tradeoff considerations involved in building automated, high-fidelity privacy safeguards for real-world AI applications, with practical strategies for implementing scalable privacy solutions in both enterprise and open-source ecosystems.