The rapid integration of artificial intelligence (AI) into cloud computing environments has introduced complex security and privacy vulnerabilities that demand rigorous evaluation. To address this, this study conducts a systematic literature review investigating the core challenges associated with deploying AI workloads across cloud architectures. While the convergence of AI and cloud infrastructures enables scalable intelligent services across diverse domains, it simultaneously introduces severe security risks, including data exfiltration, insecure Application Programming Interfaces (APIs), model extraction, adversarial attacks, and privacy inference threats. Adhering to PRISMA-inspired guidelines, relevant literature published between 2019 and 2025 was systematically harvested from major academic databases and evaluated through thematic synthesis. This review taxonomizes major security and privacy threats, synthesizes prevailing mitigation strategies, and evaluates cloud deployment architectures tailored for AI workloads. Our findings reveal that current defense mechanisms remain largely fragmented, predominantly targeting isolated technical vulnerabilities rather than providing end-to-end security integration. Furthermore, the intrinsic trade-offs among privacy preservation, system performance, scalability, and operational overhead remain inadequately resolved. Ultimately, this paper elucidates critical research gaps and outlines strategic future directions to foster the development of trustworthy, resilient, and privacy-aware AI ecosystems in cloud environments.
Copyrights © 2025