一作ICLR 2026Poster· Social Aspects / Trustworthy Machine Learning
ChatInject: Abusing Chat Templates for Prompt Injection in LLM Agents
Hwan Chang · Yonghyun Jun · Hwanhee Lee
首次获取该候选人的完整档案及 StartLux 评估共 20 积分；每位用户单独解锁，之后查看免费。主动刷新 20 积分。
实时查找电子邮箱，费用不超过20积分；未找到邮箱不扣费。
Hwan Chang · Yonghyun Jun · Hwanhee Lee
Haeun Jang · Hwan Chang · Hwanhee Lee
The deployment of Large Vision-Language Models (LVLMs) for real-world document question answering is often constrained by dynamic, user-defined policies that dictate information disclosure based on context. While ensuring adherence to these explicit constraints is critical, existing safety resear...