Evaluating the Trust and Provenance Guarantees of Multimedia Authentication Codes: a Visualization-Driven Deep Cryptanalytic Assessment of SipHash
Virendra Kumar Gautam, Dipesh Makwane, Anurag Dutta, Rajat Subhra Chakraborty · 2026
As multimedia becomes the primary vector for persuasion-optimized misinformation, the underlying authentication and provenance mechanisms that verify content origin and integrity becomes critical. Message Authentication Codes (MACs) are fundamental for safeguarding multimedia content against unauthorized modification and forgery. Differential cryptanalysis has established itself as a powerful tool for analyzing such symmetric cryptographic primitives. Recent advances in machine learning (ML) have demonstrated that ML-powered differential distinguishers can capture subtle statistical features, complementing conventional analytical techniques and offering new perspectives for security evaluation. To the best of our knowledge, this paper presents the first visualization-driven ML-based security evaluation of the widely deployed Addition-Rotation-Xor (ARX)-based MAC SipHash under chosen-message attacks. We investigate two observation domains: intermediate internal states and final authentication tags, demonstrating that the proposed distinguishers effectively capture differential characteristics across both the internal state evolution and in publicly observable outputs. The experimental results demonstrate that reduced-round SipHash preserves distinguishable statistical characteristics before complete diffusion is achieved, while classification performance progressively approaches random guessing as additional rounds are introduced. We identify ML-assisted distinguishers for SipHash-c-d, where (c, d) ∈ {(1, 0), (1, 1), (1, 2), (2, 0), (2, 1), (3, 0)}, demonstrating that recurrence-plot-based feature learning effectively evaluates the differential security of MACs. The proposed methodology strengthens the auditability of provenance infrastructure underpinning platform-scale multimedia integrity against adversarially engineered content.