AttacksarXiv 2609.16777
SAST-IR 框架：无记忆目标下多轮说服攻击成功率高达 96%
提出 SAST-IR，用有状态攻击者迭代说服无记忆模型接受反事实
- arXiv
- 2609.16777
- Published
- Layer
- Prompt layer
- Setting
- Models and APIs
- Evaluated models
- DeepSeek-Chat
- AttacksJailbreaks
- TechniqueAutomated search
Jailbreaks and attacks · Papers
- AttacksarXiv 2609.03247
研究揭示大语言模型的自发身份认证失败：五款模型的分阶段开发者身份实验
提出 MIPC 与 CFA，观察模型自拟测验后接受开发者身份
- arXiv
- 2609.03247
- Published
- Layer
- Prompt layer
- Setting
- Models and APIs
- Evaluated models
- ChatGPT, Claude, Qwen3.7-Plus and others5
- AttacksJailbreaks
- TechniqueMulti-turn escalation
Abstract自拟测验可造成对话层虚假认证，但所测授权边界并未因此改变。
这项工作把“模型自己出题、自己批改、再宣布身份已核验”定义成对话层认证失败，并与后端提权分开。
- AttacksarXiv 2605.20519
CodecAttack：在音频编解码器潜空间构造对抗扰动，绕过有损压缩防御
提出 CodecAttack，在编解码器隐空间构造可穿透有损压缩的对抗音频
- arXiv
- 2605.20519
- Published
- Layer
- Prompt layer
- Setting
- Models and APIs
- Attacker
- White-box
- Evaluated models
- Qwen2-Audio-7B-Instruct, Audio Flamingo 3, Qwen2.5-Omni
- AttacksJailbreaks
- TechniqueGradient and representation optimization
- ComponentAudio
You have reached the end