VChat: A Neural Cellular Automata-Based Cryptographically Secure Pseudo-Random Number Generator for End-to-End Encrypted Messaging

Traditional pseudo-random number generators (PRNGs) often exhibit vulnerabilities such as limited entropy, predictable patterns, and susceptibility to cryptanalytic attacks, making them unsuitable for security-critical applications. This paper presents VChat, a secure real-time messaging system powered by a novel Cryptographically Secure Pseudo-Random Number Generator (CSPRNG) based on Neural Cellular Automata (NCA). The proposed system integrates an NCA- based generator with a 2-layer Multi-Layer Perceptron (MLP) discriminator, trained through adversarial learning to produce high-entropy bit streams. The generated randomness undergoes rigorous statistical validation including ENT tests, NISTSP800-22, SP800-90B, and Dieharder bench marks, consistently achieving near-perfect entropy (7.99-8.0 bits/byte) and passing 152 out of 157NIST tests. Post-processing through HMAC- SHA512 whitening eliminates residual patterns, enabling the derivation of cryptographically secure 256-bit session keys. VChat implement send-to-end encryption with enhanced features including OTP-based authentication, Message deletion, copy functionality, and real-time profile visibility. The system ensures zero-knowledge communication where the server handles only cipher text, achieving message latency under 150ms while maintaining cryptographic strength. Experimental results demonstrate the system’s superiority over traditional PRNGs in randomness quality, security, and practical applicability for secure communication.