Artificial intelligence is becoming one of the underlying assets to contemporary cybersecurity as organizations are faced with increasing attack surfaces, scale security telemetry and the emergence of complex threats in Industry 4.0 and interconnected digital systems. The review synthesizes the benefits of AI in the fundamentals of the defence capability, including anomaly detection, intrusion detection, signature-free malware and ransomware detection, phishing and social engineering, automated incident response. It also critically evaluates the new risk environment introduced by adoption of AI and identifies adversarial threats to machine learning, model and data security threats, weaponization of AI by attackers and operational constraints of false alarms and poor generalization. Simultaneously, the paper addresses ethical issues that go hand in hand with security activities enabled by AI, such as the privacy and surveillance dilemma, the bias and discrimination of threat labeling, the transparency and explainability requirement, and the accountability of automated decision, and dual use dilemmas, governance and regulatory imperatives. The review provides a systematic taxonomy of AI applications, comparative findings about the existing evaluation procedures and data constraints, and a prospective map concerning resilient, explainable, privacy-protecting, and ethically suitable cybersecurity mechanisms that are useful in terms of long-term cyber resilience.