Selective Token-Level Cryptographic Redaction for Privacy-Preserving Clinical Deployment of Large Language Models

本文针对大型语言模型（LLM）在临床部署中因传输原始敏感健康信息而导致的隐私泄漏风险，提出了一个名为HERALD（Healthcare Encryption & Redaction via Adaptive Linguistic Decom

历史摘要 · 未重新核验 · arXiv cs.CR