用非暴力沟通约束给大模型“降火”，为对话安全提供新思路
arXiv:2606.26106v1 Announce Type: cross Abstract: Large language models (LLMs) are increasingly used in emotionally charged situations involving inter…
LLM智能体在对话开始时最脆弱，经过几次常规任务后安全性显著提升——冷启动安全差距现象首次被系统揭示。
arXiv:2606.07867v1 Announce Type: new Abstract: Are tool-calling LLM agents equally safe throughout a conversation? We discover they are not: agents a…
无需重新训练即可防御多轮越狱攻击，THRD框架精准抓住对话渐进性漏洞，保障LLM安全对话。
arXiv:2606.01738v1 Announce Type: cross Abstract: Multi-turn jailbreak attacks pose a growing threat to LLMs by exploiting conversational dynamics suc…
AI助手正在泄露你的对话？安全漏洞不容忽视，速览最新研究。
Article URL: https://leakylm.github.io/ Comments URL: https://news.ycombinator.com/item?id=48173635 Points: 1 # Comments: 0