AdvGuard framework combats LLM prompt leakage with graph convolutional networks

Researchers have developed AdvGuard, a novel framework designed to detect and mitigate system prompt leakage attacks against large language models (LLMs). Unlike previous defenses that analyze queries in isolation, AdvGuard utilizes a semantic interaction graph and graph convolutional networks to identify malicious patterns across multiple interactions. The framework includes a prompt-level detector for early protection and has demonstrated significant improvements in recall and F1 scores, substantially reducing prompt reconstruction in evaluations. AI

IMPACT Enhances LLM security by providing a robust defense against prompt leakage attacks, protecting proprietary information and safety policies.

RANK_REASON The cluster describes a research paper detailing a new method for detecting and mitigating security vulnerabilities in LLMs. [lever_c_demoted from research: ic=1 ai=1.0]

Read on Hugging Face Daily Papers →

• AdvGuard
• graph convolutional network
• LeakAgent
• LLMs
• Pleak

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →