Xiangming Gu, Xiaosen Zheng, Tianyu Pang, et al. · 2024 · ICML 2024
One adversarial image in one agent’s memory spread an ‘infectious jailbreak’ through random pairwise chats until nearly all of a million simulated agents were compromised.
Atlas / Sociology of AI / Paper
Agent Smith: A Single Image Can Jailbreak One Million Multimodal LLM Agents Exponentially Fast
Contagion & Transmission
Help improve this map
Know a paper that belongs here, or think something needs correcting?
Suggest a correction ↗Contributions are reviewed before they appear.