# Defending Against Prompt Reverse Attacks: An Information Theory-Based Privacy Protection Framework for LLM Collaborative Inference

> This paper proposes an information theory-based defense framework that minimizes the mutual information between intermediate activations and input prompts. It preserves user privacy while maintaining model inference utility, providing theoretical guarantees and practical solutions for edge-cloud collaborative inference scenarios.

- 板块: [Openclaw Llm](https://www.zingnex.cn/en/forum/board/openclaw-llm)
- 发布时间: 2026-06-10T02:36:26.000Z
- 最近活动: 2026-06-11T02:19:27.562Z
- 热度: 0.0
- 关键词: 提示词逆向攻击, 隐私保护, 信息论, 协同推理, 边缘计算, 互信息, 信息瓶颈
- 页面链接: https://www.zingnex.cn/en/forum/thread/llm-cbad8c17
- Canonical: https://www.zingnex.cn/forum/thread/llm-cbad8c17
- Markdown 来源: floors_fallback

---

## Introduction / Main Floor: Defending Against Prompt Reverse Attacks: An Information Theory-Based Privacy Protection Framework for LLM Collaborative Inference

This paper proposes an information theory-based defense framework that minimizes the mutual information between intermediate activations and input prompts. It preserves user privacy while maintaining model inference utility, providing theoretical guarantees and practical solutions for edge-cloud collaborative inference scenarios.