Shubhashree Pattanayak,
Suman Sahoo,
Sanjay Kumar Sahoo,
- Assistant professor, Department of Computer Science and Engineering, Gandhi Institute of Excellent Technocrats, Bhubaneswar, Odisha, India
- Student, Department of Computer Science and Engineering, Gandhi Institute of Excellent Technocrats, Bhubaneswar, Odisha, India
- Student, Department of Computer Science and Engineering, Gandhi Institute of Excellent Technocrats, Bhubaneswar, Odisha, India
Abstract
Artificial intelligence (AI) is seamlessly woven into vital sectors, such as self-driving cars, high-speed trading systems, and defense strategies; it has triggered a counterintuitive development in advanced cyberattacks. This survey paper attempts to perform an in-depth technical analysis of the “AI attack surface.” There are threats across three main vectors. Data integrity attacks focus specifically on examining “clean-label” poisoning and backdoor injection. Model confidentiality breaches discuss the mathematics behind model inversion and membership inference attacks. Generative exploitation examines prompt injection and hallucinations in large models. This paper presents a comprehensive technical survey of the dual nature of AI: its potential to both bolster security and be exploited as an attack vector. We first catalog and analyze the principal threat models associated with AI systems, including adversarial examples, model inversion, data poisoning, and inference attacks. We then explore privacy vulnerabilities stemming from training data leakage, unauthorized model access, and collaborative learning paradigms like federated learning. Through this lens, the survey examines how traditional threats evolve in AI contexts and highlights new attack vectors unique to learning-based systems. Next, we systematically review state-of-the-art countermeasures across defensive categories—robust training, certified defenses, differential privacy, cryptographic approaches, and secure multi-party computation—emphasizing both strengths and limitations. Fast gradient sign method (FGSM) and projected gradient descent (PGD) are examples. In addition, we investigate the combined realms of AI and the Internet of Things (IoT) using the ZIRCON framework. Instead, we promote a need for a “zero-watermarking” approach. In closing, we have a strategic outlook for 2025–2075. In fact, we believe that “explainability” or XAI is not just a norm for regulation. Rather, it is a key to unlocking a future full of artificial general intelligence.
Keywords: Adversarial machine learning, data poisoning, differential privacy, Explainable AI (XAI), Internet of Things (IoT) security, large language models, prompt injection, ZIRCON
[This article belongs to Journal of Operating Systems Development & Trends ]
References
- Paracha A, Arshad J, Farah MB, Ismail K. Machine learning security and privacy: a review of threats and countermeasures. EURASIP J Inf Secur. 2024;2024(1):10. doi:10.1186/s13635-024-00158-3.
- Mohanty SS, Tripathy S. Application of different filtering techniques in digital image processing. J Phys Conf Ser. 2021;2062(1):012007. doi:10.1088/1742-6596/2062/1/012007.
- Patil P, Chaudhary N, Prasad S, Bhandwal M, Arora M, Singh G. Predicting student performance with machine learning algorithms. 2023 3rd International Conference on Technological Advancements in Computational Sciences (ICTACS), Tashkent, Uzbekistan, 2023. p. 1346–1353. doi:10.1109/ICTACS59847.2023.10390077.
- Xue B, Zhang M, Browne WN. Particle swarm optimization for feature selection in classification: a multi-objective approach. IEEE Trans Cybern. 2013;43(6):1656–1671. doi:10.1109/TSMCB.
2227469. - Shoib S, Siddiqui MF, Turan S, Chandradasa M, Armiya’u AY, Saeed F, et al. Artificial intelligence, machine learning approach and suicide prevention: a qualitative narrative review. Prev Med Res Rev. 2025;2(6):289–297. doi:10.4103/PMRR.PMRR_121_24.
- Wang T, Zhang Y, Qi S, Zhao R, Xia Z, Weng J. Security and privacy on generative data in AIGC: a survey. ACM Comput Surv. 2024;57(4):1–34. doi:10.1145/3703626.
- Zhang Y, Zeng D, Luo J, Xu Z, King I. A survey of trustworthy federated learning with perspectives on security, robustness and privacy. In: Ding Y, Tang J, Sequeda J, et al., editors. Companion Proceedings of the ACM Web Conference 2023 (WWW ’23 Companion); 2023 Apr 30–May 4; Austin, TX, USA. New York (NY): Association for Computing Machinery; 2023. p. 1167–1176. doi:10.1145/3543873.3587681.
- Rupanetti D, Kaabouch N. Combining edge computing-assisted internet of things security with artificial intelligence: applications, challenges, and opportunities. Appl Sci. 2024;14(16):7104. doi:10.3390/app14167104.
- Wang H, Lv T, Cao Y, Li W, Zeng J, Huang P, et al. Navigating the dual-use nature and security implications of reconfigurable intelligent surfaces in next-generation wireless systems. IEEE Commun Surv Tutor. 2026;28:3346–3387. doi:10.1109/COMST.2025.3621610.
- Prajapati T. Securing the AI ecosystem: a deep dive into mobile threats, countermeasures, and privacy-preserving techniques. Artif Intell Mob Comput. 2024;1(1).
- Liu X, Qian C, Hatcher WG, Xu H, Liao W, Yu W. Secure Internet of Things (IoT)-based smart-world critical infrastructures: survey, case study and research opportunities. IEEE Access. 2019;7:79523–79544. doi:10.1109/ACCESS.2019.2920763.
- Aouedi O, Vu TH, Sacco A, Nguyen DC, Piamrat K, Marchetto G, et al. A survey on intelligent Internet of Things: applications, security, privacy, and future directions. IEEE Commun Surv Tutor. 2025;27(2):1238–1292. doi:10.1109/COMST.2024.3430368.
Journal of Operating Systems Development & Trends
| Volume | 13 | |
| Issue | 01 | |
| Received | 27/01/2026 | |
| Accepted | 31/01/2026 | |
| Published | 20/03/2026 | |
| Publication Time | 52 Days |