Generative AI in Cloud Computing: Applications, Security Challenges, and Future Research Directions

This paper reviewed the literature on Generative AI in cloud computing. The reviewed evidence shows demonstrated uses in cloud incident analysis, code assistance, and retrieval-based access to knowledge, with agent-based uses still emerging (RQ1). Recurring risks include prompt injection, sensitive-data disclosure, retrieval-data leakage, and excessive permissions (RQ2). Mitigations reduce either the likelihood or the consequences of attacks, but evaluations are limited and no single control is sufficient (RQ3). The main gaps concern cloud-specific security evaluation, agent authorization, and joint assessment of security, performance, and cost, which define the research directions in Section 9 (RQ4). For cloud service providers, the findings point to secure defaults for retrieval, permissions, and monitoring; for users and organizations, to layered controls and clear data governance. The study has limitations: it is a preliminary narrative review based on web searches rather than a systematic database search, it covers 21 sources, many results depend on specific models and versions that change quickly, and no experiments were conducted. Future work should extend the search systematically and test the proposed directions empirically.

[1] P. Mell and T. Grance, "The NIST definition of cloud computing," Nat. Inst. Standards Technol., Gaithersburg, MD, USA, NIST Special Publication 800-145, Sep. 2011, doi: 10.6028/NIST.SP.800-145. [2] A. Vaswani et al., "Attention is all you need," in Proc. Adv. Neural Inf. Process. Syst. (NeurIPS), vol. 30, 2017. [3] T. B. Brown et al., "Language models are few-shot learners," in Proc. Adv. Neural Inf. Process. Syst. (NeurIPS), vol. 33, 2020, pp. 1877–1901. [4] D. Patel et al., "Cloud platforms for developing generative AI solutions: A scoping review of tools and services," 2024, arXiv:2412.06044 (preprint). [5] P. Lewis et al., "Retrieval-augmented generation for knowledge-intensive NLP tasks," in Proc. Adv. Neural Inf. Process. Syst. (NeurIPS), vol. 33, 2020, pp. 9459–9474. [6]  T. Ahmed, S. Ghosh, C. Bansal, T. Zimmermann, X. Zhang, and S. Rajmohan, "Recommending root-cause and mitigation steps for cloud incidents using large language models," in Proc. IEEE/ACM Int. Conf. Softw. Eng. (ICSE), Melbourne, Australia, May 2023, arXiv:2301.03797. [7] Y. Chen et al., "Automatic root cause analysis via large language models for cloud incidents," in Proc. Eur. Conf. Comput. Syst. (EuroSys ’24), Athens, Greece, Apr. 2024, doi: 10.1145/3627703.3629553. [8] H. Pearce, B. Ahmad, B. Tan, B. Dolan-Gavitt, and R. Karri, "Asleep at the keyboard? Assessing the security of GitHub Copilot’s code contributions," Commun. ACM, vol. 68, no. 2, pp. 96–105, Feb. 2025. [9] K. Greshake, S. Abdelnabi, S. Mishra, C. Endres, T. Holz, and M. Fritz, "Not what you’ve signed up for: Compromising real-world LLM-integrated applications with indirect prompt injection," in Proc. ACM Workshop Artif. Intell. Secur. (AISec), Nov. 2023, arXiv:2302.12173. [10]  OWASP Foundation, "OWASP Top 10 for LLM applications 2025," Version 2025, Nov. 2024. [Online]. Available: https://genai.owasp.org/