© MAR 2025 | IRE Journals | Volume 8 Issue 9 | ISSN: 2456-8880 IRE 1707599 ICONIC RESEARCH AND ENGINEERING JOURNALS 1030 Adversarial AI and Cybersecurity: Defending Against AI- Powered Cyber Threats SHOEB ALI SYED University of the Cumberlands Abstract- The swift incorporation of Artificial Intelligence (AI) into cyber security makes a difference in digital defense systems, enabling automatic threat detection, real-time anomaly detection, and predictive analysis. Cybercriminals turned to adversarial AI approaches for developing AI weapons to compromise, evade or deceive AI- based security models. They include maliciously constructed inputs or manipulation techniques exploiting vulnerabilities of machine learning algorithms, allowing an attacker to bypass the security mechanisms, stealthily execute cyber attacks, and even corrupt AI-driven decision-making systems, among others. That makes a fierce competition for the organization to maintain robust security infrastructures within the growing arms race AI-versus-AI. This research attempts to examine how adversarial AI threats evolve and their influences on cyber security and the most effective techniques of defense against them. The research classifies the adversarial AI threat into five main types: evasion attacks, poisoning attacks, model inversion, AI-generated phishing, and adversarial malware, demonstrating their real-world instances through such studies as Deep Locker, adversarial deep fakes, and self-learning ransom ware. A mixed- method approach involved a survey of 300 cyber security professionals regarding their level of awareness of such threats and the efficacy of the defense mechanisms, namely, adversarial training, AI-enhanced intrusion detection systems, and anomaly detection algorithms. Therefore, the study implies that despite advanced AI-driven security systems being established, they may also be evaded by advanced adversarial AI attacks, necessitating proactive defenses incorporating adversarial training, AI-powered anomaly detection, and strong legal policies. It further emphasizes the immediate need for organizations to invest in and continuously monitor, share threat intelligence, and create ethical AI governance frameworks to counter adversarial attacks. Without agile, self-learning security frameworks, AI-powered defenses will remain vulnerable to sophisticated cyber attacks that adapt and evolve in real time. This paper has emerged within that developing discourse on AI cyber security by urgently advocating for engendering more resilient AI-driven security solutions against adversarial threats. The findings of the research afford valuable recommendations to cyber security experts, policymakers, and AI developers to prevent the weaponization of AI as an agent of cybercriminal exploits rather than leave it as a force for cyber security resilience. Indexed Terms- Adversarial AI, Cybersecurity, AI- Powered Cyber Threats, Evasion Attacks, Poisoning Attacks, Model Inversion, AI-Generated Phishing, Adversarial Malware, AI in Cyber Defense, Machine Learning Security, AI-Enhanced Intrusion Detection, Anomaly Detection I. INTRODUCTION 1.1 Background The intersection of Artificial Intelligence and cybersecurity revolutionizes digital security with automated threat detection, real-time anomaly detection, and predictive analytics. However, while AI fortifies the defenses against cybercriminals, these very criminals now turn their guns on AI and use it to develop adversarial AI attacks, rendering digital systems more vulnerable than ever. Adversarial AI refers to the various tactics run against a machine- learning model by manipulating its inputs, corrupting its data, or tricking an AI-backed security system. With AI security tools such as intrusion detection systems, malware detection algorithms, and automated incident response tools being increasingly seen as standard, the attacker strategies to evade, corrupt, or manipulate them are also evolving. In contrast to