GAN-Based Adversarial Encryption for Autonomous AI-Learned Cryptography
Praveen Kumar Idamakanti · International Journal on Science and Technology · 2025
GAN-based adversarial encryption leverages Generative Adversarial Networks (GANs) to enable AI agents, typically named Alice (encryptor), Bob (decryptor), and Eve (eavesdropper), to learn encryption and decryption through an adversarial game. This approach allows for the autonomous development of cryptographic protocols without explicit programming of algorithms. Advancements include integrating Genetic Algorithms (GAs) with GANs (GA-GAN) to evolve more robust and complex encryption schemes, achieving properties like perfect secrecy (One-Time Pad) under strong adversarial conditions, and extending these principles to asymmetric key encryption. The GA-GAN approach, through co-evolution of generator and discriminator networks, shows promise for developing quantum-resistant cryptography by creating dynamic, non-static encryption methods.