Prompt-in-Content Attacks: Exploiting Uploaded Inputs to Hijack LLM Behavior
This addresses a security threat in real-world LLM applications that accept user-submitted content, such as summarization tools, by exposing a subtle attack vector.
The paper identifies prompt-in-content attacks, where adversarial instructions hidden in user inputs manipulate LLM outputs, demonstrating their feasibility across platforms and analyzing root causes like prompt concatenation.
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.