AgentWatcher: A Rule-based Prompt Injection Monitor

TL;DR

AgentWatcher is a rule-based monitor that improves prompt injection detection in LLMs by focusing on causally influential context segments and providing explainable decisions, scalable to long contexts.

Contribution

It introduces a scalable, rule-based prompt injection detection method that enhances explainability and effectiveness in long-context scenarios.

Findings

Effectively detects prompt injection in tool-use agents.

Maintains utility on clean inputs without attacks.

Scales well with increasing context length.

Abstract

Large language models (LLMs) and their applications, such as agents, are highly vulnerable to prompt injection attacks. State-of-the-art prompt injection detection methods have the following limitations: (1) their effectiveness degrades significantly as context length increases, and (2) they lack explicit rules that define what constitutes prompt injection, causing detection decisions to be implicit, opaque, and difficult to reason about. In this work, we propose AgentWatcher to address the above two limitations. To address the first limitation, AgentWatcher attributes the LLM's output (e.g., the action of an agent) to a small set of causally influential context segments. By focusing detection on a relatively short text, AgentWatcher can be scalable to long contexts. To address the second limitation, we define a set of rules specifying what does and does not constitute a prompt injection, and use a monitor LLM to reason over these rules based on the attributed text, making the detection decisions more explainable. We conduct a comprehensive evaluation on tool-use agent benchmarks and long-context understanding datasets. The experimental results demonstrate that AgentWatcher can effectively detect prompt injection and maintain utility without attacks. The code is available at https://github.com/wang-yanting/AgentWatcher.