

Your agent's context is an attack surface
IA
Journalcast di Acidalia
E6
12 set 2026
28:33
Note sull'episodio
The agent harness as an attack surface: context assembly, the network path, crash recovery, and a defence that holds by construction.
- Zichuan Li, Jian Cui, Ashley Chen, Xiaojing Liao, and Luyi Xing, "What's in Your Agent's Context? Context Privilege Escalation Attacks against AI Agent Harness," arXiv:2609.01222 (2026). https://arxiv.org/abs/2609.01222
- Hanzhi Liu, Chaofan Shou, Hongbo Wen, Yanju Chen, Ryan Jingyang Fang, and Yu Feng, "Your Agent Is Mine: Measuring Malicious Intermediary Attacks on the LLM Supply Chain," arXiv:2604.08407 (2026). https://arxiv.org/abs/2604.08407
- Guanlong Wu, Dahui Li, Ke Jiang, Jianyu Niu, Cong Wang, and Yinqian Zhang, "Safe to Resume? Breaking Execution Continuity of Agent Execution via Rollback," arXiv:2608.29381 (2026). https://arxiv.org/abs/2608.29381
- Edoardo Debenedetti, Ilia Shumailov, Tianqi Fan, Jamie Hayes, Nicholas Carlini, Daniel Fabian, Christoph Kern, Chongyang Shi, Andreas Terzis, and Florian Tramer, "Defeating Prompt Injections by Design," arXiv:2503.18813 (2025). https://arxiv.org/abs/2503.18813
Parole chiave
language models
AI safety