Threat Landscape in Artificial Intelligence Systems: Taxonomy, Attack Vectors and Security Implications

Abstract

The rapid integration of Artificial Intelligence (AI) systems across critical sectors such as healthcare, finance, autonomous transportation, and national security has fundamentally altered the global cybersecurity threat landscape. Unlike traditional software systems, AI introduces novel vulnerabilities rooted in data-driven learning, model opacity, and high dimensional decision boundaries. This paper presents a comprehensive analysis of the evolving threat landscape in AI systems, focusing on adversarial machine learning attacks, data poisoning, privacy inference, model extraction, supply-chain vulnerabilities, and emerging risks in generative AI and large language models (LLMs). A structured taxonomy of AI-specific threats is proposed, mapping attack vectors to lifecycle stages and adversary capabilities. The study further evaluates real world attack scenarios, sector specific impacts, and systemic risks arising from interconnected AI ecosystems. The paper concludes by outlining detection strategies, governance considerations, and future research directions necessary to ensure secure, trustworthy, and resilient AI deployments