QueryIPI : Query-agnostic Indirect Prompt Injection on Coding Agents
Yuchong Xie 1 ∗ Zesen Liu 1 ∗ Mingyu Luo 2 Zhixiang Zhang 1 Kaikai Zhang 1 Zongjie Li 1 Ping Chen 2 Shuai Wang 1 Dongdong She 1 † 1 The Hong Kong University of Science and Technology 2 Fudan University · 2025
Read QueryIPI : Query-agnostic Indirect Prompt Injection on Coding Agents by Yuchong Xie 1 ∗ Zesen Liu 1 ∗ Mingyu Luo 2 Zhixiang Zhang 1 Kaikai Zhang 1 Zongjie Li 1 Ping Chen 2 Shuai Wang 1 Dongdong She 1 † 1 The Hong Kong University of Science and Technology 2 Fudan University (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-23675.txt.
Description
Yuchong Xie and nine co-authors expose a new attack on AI coding assistants: indirect prompt injection that fires no matter what the user asks. Coding agents inside editors wield powerful tools and deep system access, so a hidden malicious instruction is a serious risk. Earlier injections needed one particular user query as a trigger, which made them unreliable in the wild.
The team's framework, QueryIPI, hides the payload inside tool descriptions instead — text the agent always sees, unlike the changing user query. An automated loop refines those descriptions through black-box trial and error, starting from seeds shaped to fit agent conventions and reflecting on failures and refusals. Across five simulated agents, the attack succeeded up to eighty-seven percent of the time, against fifty percent for the best prior method. Worse, the crafted descriptions transferred to real-world coding agents, marking a practical threat.
Catalog record
| CREATOR | Yuchong Xie 1 ∗ Zesen Liu 1 ∗ Mingyu Luo 2 Zhixiang Zhang 1 Kaikai Zhang 1 Zongjie Li 1 Ping Chen 2 Shuai Wang 1 Dongdong She 1 † 1 The Hong Kong University of Science and Technology 2 Fudan University |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 45 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-23675