Search papers, labs, and topics across Lattice.
This paper investigates the efficacy of prompt injection attacks against AI agents that utilize modular skills selected by routers, demonstrating that existing evaluations overestimate attack success by neglecting the skill retrieval phase. The authors introduce CORSA, a novel router-aware attack strategy that optimizes injections for both retrieval and subsequent execution across task clusters. Experiments reveal CORSA significantly boosts attack success rates while maintaining user utility and exhibiting transferability across diverse router architectures and LLM backbones.
Prompt injections are far less effective in realistic multi-skill agent settings than previously thought, but a new router-aware optimization strategy can achieve high success rates by targeting skill retrieval.
AI agents increasingly rely on modular third-party"skills"that are dynamically selected by skill routers to execute complex tasks. While recent studies highlight the threat of prompt injections embedded in these skills, existing evaluations often assume settings where the malicious skill is already selected for execution. We show that this assumption can substantially overestimate attack success. In realistic multi-skill environments, injected skills must first compete for retrieval, reducing the effective attack success rate (ASR) of existing injections by 87-97%. To address this limitation, we introduce CORSA (Cluster Optimization for Router-Aware Skill Attacks), a router-aware attack that optimizes skill injections for both retrieval and execution across clusters of related tasks. We evaluate skill injection attacks under router-managed multi-skill settings by extending the benchmark introduced by SkillRouter with eight malicious payload categories. CORSA uses successive optimization stages to first improve retrieval and then optimize end-to-end attack success, while we evaluate user utility and injection naturalism separately. Our experiments show that CORSA substantially improves both retrieval and end-to-end attack success over existing skill injections while preserving user utility, and that the resulting attacks transfer across different router architectures and LLM backbones.
