Research · curated 11 Aug 2026
AgentWatcher: A Rule-based Prompt Injection Monitor
First reported arxiv.org
Coverage timeline
Single-source research — first reported, latest, and curated coincide.
Why it matters
AgentWatcher targets a core defensive gap — detecting indirect prompt injection in long-context, tool-using agents with explainable, rule-based decisions — which is directly useful to defenders hardening deployed LLM agents.
AgentWatcher is a proposed rule-based prompt injection monitor for LLM agents that attributes a model's output to a small set of causally influential context segments and then uses a monitor LLM to reason over explicit rules about what does and does not constitute prompt injection. The authors report evaluation on tool-use agent benchmarks and long-context datasets, and publish code at github.com/Wang-Yanting/AgentWatcher.