#one-relation attack

1 stories · 1 in 7 days · 1 in 30 days

AI安全 1 LLM 1 智能体 1 论文 1

Stories, products and papers tagged "one-relation attack", newest first.

Oct 20

Papers Builders09:30

Chinese content一句话即可误导智能体推理：论文提出 one-relation attack

18 个模型里一句真话就能让智能体改错答案，开源模型中招率 80%，做 Agent 安全的必看这篇。

#LLM #智能体 #AI安全 #one-relation attack

arXiv cs.AI@Jinwei Hu et al. (4)Source