CYBERSECURITY STANDARDS FOR AGENTIC AI SYSTEMS

Abstract

This project examines the cybersecurity challenges associated with Agentic Artificial Intelligence (AI) systems, which are capable of autonomous decision-making and adaptive behaviour. It evaluates the limitations of existing cybersecurity and governance frameworks, with particular emphasis on ISO/IEC 42001, in addressing emerging AI-specific threats. The study investigates key risks including adversarial machine learning, data poisoning, model inversion, unauthorized model use, and ethical concerns such as bias and transparency. Based on these findings, the project proposes practical security enhancements, including adversarial testing, secure data pipelines, robust access controls, explainable AI techniques, continuous monitoring, and AI-specific governance policies. The report also highlights the importance of lifecycle-based security management and awareness programs to improve organizational resilience against evolving AI threats. The findings provide a foundation for developing secure, trustworthy, and responsible Agentic AI systems while supporting future technical implementation and compliance with emerging AI security standards.