Problem
Research questions and friction points this paper is trying to address.
prompt injection attacks
multi-agent robotic systems
large language models
autonomous robots
adversarial actions
When Prompts Control Robots: Prompt Injection Attacks in Multi-Agent Robotic Systems
🤖 AI Summary
This study systematically investigates prompt injection attacks against large language model (LLM)-driven multi-agent robotic systems, which are vulnerable to unsafe decision-making and even physical harm due to their susceptibility to such exploits. The work is the first to comprehensively examine both direct injections via task instructions and indirect injections through perception modules, revealing how adversarial prompts propagate across agents via shared prompting structures. It further analyzes how prompt composition and target agent characteristics influence attack efficacy. Experimental results demonstrate that prompt injections can induce adversarial behaviors, substantially degrade task completion rates, and that system architecture design critically determines attack success rates in multi-agent settings.
📝 Abstract
Large language models are increasingly integrated into autonomous robotic systems for task planning and control, but this integration exposes them to prompt injection attacks that can lead to unsafe decisions and physical harm. Multi-agent settings increase the risks through cross-agent contamination and broader attack surfaces. In this paper, we evaluate prompt injection attacks against an LLM-based multi-agent robotic system, considering both direct injections into task instructions and indirect injections through perception modules. In our experiments across varying attack-goal complexities and injection strategies in both single-agent and multi-agent settings, we show that prompt injection can induce adversarial actions while reducing task completion. We find that attacks can propagate from one agent to others through shared prompt structures, with impacts varying depending on prompt composition and the targeted agent. We further analyze how architectural changes affect LLM queries and, consequently, the attack success. To the best of our knowledge, this is the first study that systematically investigates prompt injection attacks in a multi-agent LLM-based robotic system.
Innovation
Methods, ideas, or system contributions that make the work stand out.
prompt injection
multi-agent robotic systems
large language models
adversarial attacks
cross-agent contamination
🔎 Similar Papers
No similar papers found.
