🤖 AI Summary
This work addresses the propagation and amplification of malicious instructions caused by prompt injection attacks in multi-agent systems by proposing a model-agnostic defense pipeline that integrates a nested learning architecture with a semantic caching mechanism to enable secure, efficient, and auditable mitigation. The study introduces the novel TIVS-O evaluation framework, which incorporates an observability scoring ratio to reveal a non-monotonic trade-off between security mitigation and audit transparency, and employs a five-dimensional metric suite to jointly optimize safety, performance, and sustainability. Experimental results demonstrate that the proposed approach eliminates high-risk vulnerabilities, reduces large language model invocations by 41.6%, and significantly lowers latency, energy consumption, and carbon emissions, thereby validating the feasibility of production-grade secure and green deployment.
📝 Abstract
Prompt injection remains a central obstacle to the safe deployment of large language models, particularly in multi-agent settings where intermediate outputs can propagate or amplify malicious instructions. Building on earlier work that introduced a four-metric Total Injection Vulnerability Score (TIVS), this paper extends the evaluation framework with semantic similarity-based caching and a fifth metric (Observability Score Ratio) to yield TIVS-O, investigating how defence effectiveness interacts with transparency in a HOPE-inspired Nested Learning architecture. The proposed system combines an agentic pipeline with Continuum Memory Systems that implement semantic similarity-based caching across 301 synthetically generated injection-focused prompts drawn from ten attack families, while a fourth agent performs comprehensive security analysis using five key performance indicators. In addition to traditional injection metrics, OSR quantifies the richness and clarity of security-relevant reasoning exposed by each agent, enabling an explicit analysis of trade-offs between strict mitigation and auditability. Experiments show that the system achieves secure responses with zero high-risk breaches, while semantic caching delivers substantial computational savings, achieving a 41.6% reduction in LLM calls and corresponding decreases in latency, energy consumption, and carbon emissions. Five TIVS-O configurations reveal optimal trade-offs between mitigation strictness and forensic transparency. These results indicate that observability-aware evaluation can reveal non-monotonic effects within multi-agent pipelines and that memory-augmented agents can jointly maximize security robustness, real-time performance, operational cost savings, and environmental sustainability without modifying underlying model weights, providing a production-ready pathway for secure and green LLM deployments.