EventPast event
Researchers propose a pair-aware optimization framework to jailbreak LLM agents
Start here
AI overview
2026年10月，arXiv 论文提出 Pair-aware optimization 框架，用于对 LLM 智能体实施越狱攻击：以零阶优化探索对抗前缀、一阶优化精炼对应后缀，保留多个前缀候选并在候选对之间选择。论文比较纯 ZO、纯 FO、ZO->FO 与 FO->ZO 四种策略，称混合优化在恶意指令与恶意环境两类基准上比单一优化器方法更有效，且优化顺序对性能影响显著。作者还从随机生成角度考察智能体越狱。
Generated by AI from reports · Updated 24 days ago
Attention and developments
Attention, hourly
Now 29, 0% in 24 hours·Peak 29 (Nov 6 04:00)
Nov 7Nov 8Nov 9Nov 10Nov 11Nov 12Nov 13Now 29
Related research
This event's paper
- Pair-Aware Optimization: Coordinating Exploration and Refinement for Jailbreaking LLM Agents
arXiv · 2026-10-17 · Deep reading
Progress timeline
Oct 18
- arXiv: Jailbreaks, Prompt Injection, Poisoning & DefensesPair-Aware Optimization: Coordinating Exploration and Refinement for LLM Agent Jailbreaks