A Symmetric Cipher Construction with Ambiguity-Based Security
Abstract
This paper presents a parameterized symmetric cipher construction that renders the hash output unreconstructible from ciphertext-only observation. Noise absorption interposes a CSPRNG-generated random container between the PRF hash output and the observer; each byte retains one random noise bit at an unknown position. Under known-plaintext, chosen-plaintext, and chosen-ciphertext attacks the closure is computational and PRF-conditional; the information-theoretic property scopes to the noise-absorption layer under passive observation (Theorem 1). Encoding ambiguity applies a secret rotation (0–6) from an independent per-snake dataSeed to each pixel's data bits, creating 7^P unverifiable configurations across P pixels. A mandatory, always-on 48-bit Interlocked Barrier forms a second architectural layer: each 48-bit chunk of the payload is partitioned into three disjoint 16-of-48 lane payloads by a PRF-keyed balanced mask triple drawn from a space of cardinality ≈ 2^70.20 per chunk, unobservable without the dedicated lockSeed. Even if the noise mechanism is bypassed via chosen-ciphertext attack (CCA reveals noise positions), the rotation barrier, the per-chunk mask permutation, and CSPRNG residue in data positions survive through 8-seed isolation. The 8-seed architecture (noiseSeed, lockSeed, three per-snake dataSeeds, three per-snake startSeeds) ensures that compromise of any single configuration domain provides zero information about the remaining domains; the lockSeed → per-chunk mask path is bound to the primitive through cascade PRF binding (two consecutive live PRF calls). A dual-nonce wire format carries two independently CSPRNG-drawn nonces per message, producing independent configurations per encryption with no caller-addressable override. The construction exhibits ambiguity-based security: the number of observation-consistent configurations grows exponentially with data size. This property is orthogonal to Shannon's key-entropy bound and distinct from Shannon's perfect-secrecy relationship on plaintext entropy. Above a threshold P_th = ⌈k / log₂ C⌉ (C = 56 without CCA, C = 7 under CCA), encoding ambiguity exceeds the 2^k key space. At 64 KB plaintext, ambiguity reaches 2^26,414; its exponent is 25.8× the 1024-bit key-space exponent. An attacker-realistic audit suite in the reference implementation confirms the barrier's absorption across multiple trapdoor mechanism classes at the tested sample sizes. All operations are elementary: XOR, bitwise AND, modular reduction, and bit shifts. No Galois fields, no S-boxes, no polynomial multiplication. The security architecture composes around a pluggable PRF hash function; the construction's own operations are elementary, while the closures under KPA/CPA/CCA remain computational and PRF-conditional on the primitive.
// Source
Authors: Andrey Kuvshinov