Prompt Injection Mitigation with Agentic AI, Nested Learning, and AI Sustainability via Semantic Caching

ABSTRACT

Defending autonomous systems from prompt injection remains a critical security hurdle. This paper introduces a multi-layered defense architecture utilizing a dedicated 'Security Agent' equipped with nested learning to identify malicious intent before it reaches the execution layer. Furthermore, the framework integrates semantic caching to both improve resistance against repetitive attacks and drastically reduce the carbon footprint of massive inference workloads.