最近一年的 AI 安全论文
本站共收录 79 篇- 用课程强化学习对前沿模型做提示注入红队测试
Climbing the Hill: Prompt Injection Red-Teaming Against Frontier Models with Curriculum Reinforcement Learning
- PIMiner：无需再训练即可跨模型迁移的提示注入红队智能体系统
Agent Against Agent: An Agentic System for Automatic Prompt Injection Red Teaming
- CleanBase：检测 RAG 知识库中的恶意文档
CleanBase: Detecting Malicious Documents in RAG Knowledge Databases
- FlashRT：面向长上下文 LLM 提示注入与知识投毒的高效红队框架
FlashRT: Towards Computationally and Memory Efficient Red-Teaming for Prompt Injection and Knowledge Corruption
- TRUSTDESC：通过生成可信工具描述防止 LLM 应用中的工具投毒
TRUSTDESC: Preventing Tool Poisoning in LLM Applications via Trusted Description Generation
- EnsembleSHAP：面向随机子空间方法的忠实且可认证鲁棒的特征归因方法
EnsembleSHAP: Faithful and Certifiably Robust Attribution for Random Subspace Method
- PISmith：基于强化学习的提示注入防御红队测试框架
PISmith: Reinforcement Learning-based Red Teaming for Prompt Injection Defenses
- TASO：通过模板与后缀交替优化越狱大语言模型
TASO: Jailbreak LLMs via Alternative Template and Suffix Optimization
- PISanitizer：通过提示词清洗防止长上下文大语言模型的提示注入
PISanitizer: Preventing Prompt Injection to Long-Context LLMs via Prompt Sanitization
署名以 dblp 和本人的 Google Scholar 为准；只列本站收录的 AI 安全论文。