Encrypted intelligence: A comparative analysis of homomorphic encryption frameworks for privacy-preserving AI

This paper presents a comparative study of various homomorphic encryption models to evaluate their qualitative and quantitative benefits and drawbacks in performing computations on encrypted data. Within the framework of ethical AI, the study focuses on enhancing privacy, secrecy, and security, addressing limitations in existing privacy-preserving solutions such as differential privacy and secure multi-party computation. To provide context, related encryption paradigms such as symmetric, asymmetric, hybrid, and multi-party computation are also discussed. The review synthesizes findings from recent literature, comparing schemes based on key performance metrics including encryption and decryption speed, memory consumption and quantum resistance. Published benchmark results and case studies are used to highlight trade-offs between privacy guarantees and computational feasibility. The study highlights the practicality of homomorphic encryption for real-world applications, providing information on its potential to advance privacy-preserving AI while maintaining computational feasibility. The paper also surveys practical applications of homomorphic encryption in machine learning, secure data analytics, and federated learning, along with emerging challenges such as quantum-safe cryptography and hardware acceleration. This review serves as a consolidated reference for researchers and practitioners seeking to select appropriate encryption techniques for AI applications, providing both a broad overview of the field and a focused discussion on homomorphic encryption’s capabilities and limitations.