Evaluating Genetic-Algorithm Key Generation for OTP Encryption Against AEAD Standards

One-Time Pad (OTP) encryption offers perfect secrecy, but only if its key is truly random, as long as the message, and never reused. Prior studies proposed heuristic key generators for OTP, such as genetic algorithms (GA), linear congruential generators (LCG) and neural networks, usually without a unified comparison against the authenticated-encryption (AEAD) standards used in practice. This paper presents a reproducible benchmark of five methods: OTP-OS, LCG-XOR, GA-XOR, AES-GCM and ChaCha20-Poly1305. The main experiment contains 1,200 randomized observations over four message sizes (4 KiB to 5 MiB) with 30 replicated blocks, complemented by tamperdetection, keystream-diagnostic and GA-isolation experiments. AES-GCM was the fastest method from 64 KiB upward (6.90 ms at 5 MiB), whereas GA-XOR was 17.4 to 1,554.9 times slower than the fastest method depending on size. Each AEAD cipher rejected all 90 tampered packages, while each of the three XOR-based methods accepted all 90 modifications silently. LCG-XOR reached an ideal entropy of 8.000000 bit/byte yet repeated its keystream every 256 bytes. Finally, GA improved its own fitness function over a single seed but not over random search with an equal budget of 156 evaluations, nor on held-out data. Heuristic key generation therefore showed no demonstrated security benefit over standard AEAD.