Hongwei Li
University of Electronic Science and Technology of China
最近一年的 AI 安全论文
本站共收录 34 篇- FragToken：通过非规范 Token 生成放大 LLM 推理成本
FragToken: Amplifying LLM Inference Costs through Noncanonical Token Generation
- RAPID：面向文生图服务的实时防未授权蒸馏防御
RAPID: A Real-Time Defense Against Unauthorized Model Distillation for Text-to-Image Services
- CounterPersona：在仅追加场景下防御未经授权的人格技能蒸馏
CounterPersona: Append-Only Defense Against Unauthorized Persona Skill Distillation
- EvoTrustRAG：面向可靠检索增强生成的演化感知冲突归因与证据处理
EvoTrustRAG: Evolution-Aware Conflict Attribution and Evidence Handling for Reliable Retrieval-Augmented Generation
- InkShield：抵御未经授权手写模仿的书写风格保护
InkShield: Writing Style Protection Against Unauthorized Handwriting Mimicry
- 针对专有 LLM Agent 的黑盒技能窃取攻击：一项实证研究
Black-Box Skill Stealing Attack from Proprietary LLM Agents: An Empirical Study
- MLJailDe：学习语言无关意图表示以检测多语言越狱
One Jailbreak, Many Tongues: Learning Language-Insensitive Intention Representations for Multilingual Jailbreak Detection
- （误）对齐的艺术：微调方法如何在后训练中有效误对齐与再对齐大语言模型
The Art of (Mis)alignment: How Fine-Tuning Methods Effectively Misalign and Realign LLMs in Post-Training
- BadTemplate：通过 chat template 对大语言模型的免训练后门攻击
BadTemplate: A Training-Free Backdoor Attack via Chat Template Against Large Language Models
- State Backdoor：面向视觉-语言-动作（VLA）模型状态空间的隐蔽真实世界投毒攻击
State Backdoor: Towards Stealthy Real-world Poisoning Attack on Vision-Language-Action Model in State Space
- TEAR：针对文生视频模型的时间感知自动化红队测试框架
TEAR: Temporal-aware Automated Red-teaming for Text-to-Video Models
署名以 dblp 和本人的 Google Scholar 为准；只列本站收录的 AI 安全论文。