With the rapid growth of cyber threats in digital infrastructure, Artificial Intelligence-driven Intrusion Detection Systems (AI-IDS) have emerged as a crucial innovation in cybersecurity. While individual studies highlight the benefits of various AI-based detection models, there remains a limited holistic analysis of their algorithmic foundations, domain-specific applications, and ethical implications. This study aims to conceptualise the role of AI-IDS across multiple network environments and assess their performance in modern cyber defence. Furthermore, it identifies emerging opportunities while critically analysing the ethical considerations. This study adopts the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) protocol to ensure methodological rigor. A comprehensive search was conducted across IEEE Xplore, SpringerLink, Science Direct, and Google Scholar. Following a multi-stage screening process, only 36 studies met the inclusion criteria. The results reveal that AI-IDS significantly improves threat detection accuracy, particularly in identifying both known and zero-day attacks, while reducing false positives and enhancing real-time responsiveness. Application domains include enterprise networks, IoT environments, cloud platforms, and critical infrastructure systems. Deep learning, ensemble models, and hybrid approaches were prominent techniques in high-performing systems. The review uncovered critical gaps, including insufficient standardization in evaluation metrics, limited cross-platform adaptability, and inadequate analysis of long-term efficacy under evolving threats. These insights are valuable for policymakers, cybersecurity professionals, and researchers working to safeguard increasingly complex digital ecosystems.