Ignored, Not Refused: What Low Prompt-Injection Success Rates Hide in Small Tool-Using Agents
The paper evaluates the effectiveness of benchmarks for indirect prompt injection security in tool-using agents and proposes a new evaluation metric.
The paper proposes a new evaluation metric, the Attack-Task Outcome Matrix (ATOM), to better measure the effectiveness of defenses against indirect prompt injection attacks.
Before reading this…
Applications
- →Improving the security of tool-using agents
To understand this paper, make sure you know these concepts first:
- Understanding of tool-using agents and indirect prompt injection attacksfind papers →
Abstract
More Like ThisTool-using agents read untrusted external content, which exposes them to indirect prompt injection (IPI), where hidden instructions hijack the agent's tool calls. Benchmarks measure IPI security by the attack success rate (ASR), which treats every failed attack as a successful defense. To test this assumption, we audit 33,000 trajectories from eleven open-weight agents (2-14B parameters) on InjecAgent and AgentDojo. A label-blind LLM judge shows that most failed attacks are ignored (76-80%) rather than refused (17-18%). With identical model weights, changing only the system prompt shifts InjecAgent ASR from 0.2% to 43%. Our Attack-Task Outcome Matrix (ATOM) further shows that in 21.7% of AgentDojo episodes, the agent stayed safe but did not complete the user's task. A low ASR therefore mostly reflects agents silently ignoring injections, not refusing them. Evaluations should report how attacks fail and whether the user's task is completed, alongside ASR. We release the labeled corpus, judge prompt, and analysis code.