一句话：对 GPT 等 LLM 应用平台做系统化威胁建模，提炼三类威胁模型与 26 个攻击向量，19 个经真实验证。
随着大语言模型（LLM）的普及，基于 LLM 的定制应用（如 OpenAI 的 custom GPTs）迅速增多，它们提供网页浏览、代码执行等功能，同时也引入了严重的安全与隐私威胁。本文对 GPT 平台引发的安全与隐私问题展开系统分析：依据恶意行为者的角色，将潜在攻击场景归纳为三类威胁模型，并识别出 GPT 中的关键数据交换通道；随后借用 STRIDE 威胁建模框架，系统枚举该生态中的攻击面，共得到 26 个潜在攻击向量，其中 19 个在真实环境中被部分或完全验证。研究结果表明，GPT 生态及其它类似的 LLM 应用生态亟需更健壮的安全与隐私防护措施。
The emergence of large language models (LLMs) has significantly accelerated the development of a wide range of applications across various fields. There is a growing trend in the construction of specialized applications based on LLMs, such as the custom GPTs by OpenAI. While custom GPTs provide various functionalities like web browsing and code execution, they also introduce significant security threats. In this paper, we conduct a comprehensive analysis of the security and privacy issues arising from the GPT platform. Our systematic examination categorizes potential attack scenarios into three threat models based on the role of the malicious actor, and identifies critical data exchange channels in GPTs. Utilizing the STRIDE threat modeling framework, we identify 26 potential attack vectors, with 19 being partially or fully validated in real-world settings. Our findings emphasize the urgent need for robust security and privacy measures in the GPT ecosystem as well as other similar LLM application ecosystems.
@inproceedings{paperbot9127,
title = {A Systematic Threat Modeling of LLM Applications},
author = {Guanhong Tao and Siyuan Cheng and Zhuo Zhang and Junmin Zhu and Guangyu Shen and Wanjing Han and Mu Zhang and Xiangyu Zhang},
booktitle = {FSE},
year = {2025},
doi = {10.1145/3696630.3728712}
}