Building Trust in Intelligent Systems: An AI-Based Intrusion Detection System for Detecting AI-Generated Cyber Attacks

Abstract

The rapid evolution of Artificial Intelligence (AI) has transformed the cyber threat landscape, enabling attackers to generate sophisticated and automated attacks that bypass traditional security mechanisms. This study proposes a holistic multimodal AI-Based Intrusion Detection System (AI-IDS) aimed at strengthening trust in intelligent systems by effectively detecting AI-generated cyber-attacks across three dimensions: network traffic, malicious URLs, and phishing emails. The proposed framework integrates an XGBoost-based Network Intrusion Detection System trained on the CSE-CIC-IDS2018 dataset with SMOTE-based balancing, an XGBoost-driven malicious URL detection module using lexical features, and a hybrid CNN-LSTM model for identifying AI-generated phishing emails through semantic analysis. Experimental results demonstrate strong performance, achieving 97% accuracy for network intrusion detection, 90.71% accuracy for malicious URL detection, and 99.1% accuracy with an AUC of 0.997 for phishing email detection. A Streamlit-based prototype secured via Cloudflare Tunnel illustrates practical deployment within a Zero Trust environment.