IAGA Case Study: End-to-End Reliability and Security Benchmark

This document presents the results of an end-to-end benchmark conducted on the IAGA (Intelligent API Gateway & Armor) platform, a security and reliability gateway designed for applications built on Large Language Models. The evaluation was carried out using a purpose-built dataset of 297 requests, spanning legitimate traffic, targeted adversarial attacks, prompts containing personally identifiable information (PII), and semantically paraphrased queries for cache evaluation. The benchmark demonstrates that IAGA achieves a 100% attack block rate across prompt injection, SQL injection, and jailbreak categories, while maintaining a false positive rate of only 0.51%. The PII redaction module correctly identified and masked sensitive data in all tested prompts. Semantic caching yielded a 22% hit rate on paraphrased queries during warm runs, producing a 1.8x latency reduction on cache hits. These findings confirm the viability of IAGA as a transparent proxy layer that adds security, privacy protection, and performance optimization to LLM-based application stacks without requiring modifications to the underlying provider or client code.