Get every episode summarized
Each time Seventy3 publishes, we email you a written briefing from the transcript — the topics, who appeared, and any specific claims, with the ad reads skipped.
Email me new episodesFree for 3 shows. No card needed.
About this episode
Seventy3：借助NotebookLM的能力进行论文解读，专注人工智能、大模型、机器人算法方向，让大家跟着AI一起进步。
今天的主题是：
From Prompt Injections to Protocol Exploits: Threats in LLM-Powered AI Agents Workflows
Summary
本论文深入探讨了大型语言模型（LLM）驱动的AI代理生态系统面临的复杂安全威胁。它提出了一个统一的端到端威胁模型，涵盖了从主机到工具以及代理间通信的各个方面。作者详细分类并审查了超过三十种攻击技术，包括输入操纵（如提示注入）、模型泄露（如后门和数据中毒）、系统和隐私攻击（如侧信道和成员推断），以及协议漏洞。该研究还评估了现有防御措施的有效性，并指出了未来的研究方向和关键挑战，旨在指导鲁棒防御机制的设计和安全最佳实践的建立，以确保LLM代理工作流的弹性。
原文链接：https://arxiv.org/abs/2506.23260
前往小宇宙评论区与主播互动
Get every episode summarized
Each time Seventy3 publishes, we email you a written briefing from the transcript — the topics, who appeared, and any specific claims, with the ad reads skipped.
Email me new episodesFree for 3 shows. No card needed.
Know when Jiajun Zhu turns up
Follow Jiajun Zhu and once a week we email you every new episode they appeared on — including guest spots the show notes never mention, because we read the transcript.
Follow Jiajun ZhuFree. Pick your own day and time.
Hosts & guests
No transcript yet
This episode has not been transcribed. Request it and it moves to the front of the queue.
More episodes
