Event ·Official first-hand

Self-Evolving Defense:LLM智能体安全框架

研究人员提出Self-Evolving Defense (SED)框架,无需重新训练模型即可持续学习安全策略。该框架在DeepSeek V4 Flash、GLM 5.2和Kimi K3三个开源模型上测试,在AGENTDOJO基准上将目标提示注入成功率降至0.42%,而最佳基线防御为3.7%。在HARMBENCH基准上,SED将X-TEAMING攻击成功率控制在7.8%,远低于最佳基线的35.2%,同时保持良性任务效用。

Current read

SED框架让LLM智能体无需重训练就能持续防御新型攻击,在DeepSeek、GLM和Kimi模型上都表现出色。

8 sources63° AI heatLast updated 2026/9/29 03:19:11

Evidence

8 sources
Primary sourcearXiv: DeepSeek
View source
Related source 2Xianbao Qian
View source
Related source 6Fireworks AI
View source
Related source 7Aravind Srinivas
View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.