Systematic literature analysis of artificial intelligence driven penetration testing for ARM systems

The rapid convergence of artificial intelligence and cybersecurity presents significant opportunities for automated penetration testing that can discover, prioritize, and remediate vulnerabilities at machine speed. However, deployment on resource-constrained ARM platforms remains largely unexplored. This systematic literature analysis synthesizes 38 peer-reviewed studies (2019–2025) using PRISMA 2020 to evaluate AI-driven penetration testing techniques and their ARM deployment feasibility. The four major AI paradigms analyzed include: (1) traditional machine learning achieving 85–99% accuracy with minimal resource footprint, (2) deep learning approaches requiring significant optimization for 95 MB deployment targets, (3) large language models facing fundamental 100x memory mismatches, and (4) reinforcement learning demonstrating 78–92% autonomous success rates. The analysis reveals that while individual AI techniques show remarkable laboratory performance, only 23% of research addresses ARM platform constraints. The critical deployment barriers include: memory requirements ranging from 10 MB to 80 GB, architectural variance impacting inference performance by 50%, and thermal limitations that cause 30% performance degradation. This analysis synthesizes existing research and identifies critical gaps toward developing edge-compatible security solutions, charting key research directions for deploying AI practically across the 280+ billion ARM devices operating globally.