Prompt-in-Content Attacks: Exploiting Uploaded Inputs to Hijack LLM Behavior

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Lian, Zhuotao, Wang, Weiyu, Zeng, Qingkui, Nakanishi, Toru, Kitasuka, Teruaki, Su, Chunhua
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908505217171456
author Lian, Zhuotao
Wang, Weiyu
Zeng, Qingkui
Nakanishi, Toru
Kitasuka, Teruaki
Su, Chunhua
author_facet Lian, Zhuotao
Wang, Weiyu
Zeng, Qingkui
Nakanishi, Toru
Kitasuka, Teruaki
Su, Chunhua
contents Large Language Models (LLMs) are widely deployed in applications that accept user-submitted content, such as uploaded documents or pasted text, for tasks like summarization and question answering. In this paper, we identify a new class of attacks, prompt in content injection, where adversarial instructions are embedded in seemingly benign inputs. When processed by the LLM, these hidden prompts can manipulate outputs without user awareness or system compromise, leading to biased summaries, fabricated claims, or misleading suggestions. We demonstrate the feasibility of such attacks across popular platforms, analyze their root causes including prompt concatenation and insufficient input isolation, and discuss mitigation strategies. Our findings reveal a subtle yet practical threat in real-world LLM workflows.
format Preprint
id arxiv_https___arxiv_org_abs_2508_19287
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Prompt-in-Content Attacks: Exploiting Uploaded Inputs to Hijack LLM Behavior
Lian, Zhuotao
Wang, Weiyu
Zeng, Qingkui
Nakanishi, Toru
Kitasuka, Teruaki
Su, Chunhua
Cryptography and Security
Artificial Intelligence
Large Language Models (LLMs) are widely deployed in applications that accept user-submitted content, such as uploaded documents or pasted text, for tasks like summarization and question answering. In this paper, we identify a new class of attacks, prompt in content injection, where adversarial instructions are embedded in seemingly benign inputs. When processed by the LLM, these hidden prompts can manipulate outputs without user awareness or system compromise, leading to biased summaries, fabricated claims, or misleading suggestions. We demonstrate the feasibility of such attacks across popular platforms, analyze their root causes including prompt concatenation and insufficient input isolation, and discuss mitigation strategies. Our findings reveal a subtle yet practical threat in real-world LLM workflows.
title Prompt-in-Content Attacks: Exploiting Uploaded Inputs to Hijack LLM Behavior
topic Cryptography and Security
Artificial Intelligence
url https://arxiv.org/abs/2508.19287