What question did this study set out to answer?

To develop a framework that uses causal inference for analyzing security margins and identifying cipher architectures.

February 12, 20260 citationsOpen Access

.causal Cryptanalysis: Black-Box Security Margins and Blind Cipher Identification

Key Points

To develop a framework that uses causal inference for analyzing security margins and identifying cipher architectures.
Implemented deterministic causal graph inference on cipher outputs.
Introduced the Causal Amplification Security Index (CASI) for quantifying security weakness.
Analyzed six cipher families spanning four architectural classes without the need for training or tuning.
CASI detection frontiers align with known cryptanalysis results for six cipher families.
Achieved 100% successful classification of five architecture classes from reduced-round outputs.
Matched Gohr's neural distinguisher for the Speck 32/64 cipher within one round of analysis.

Abstract

We demonstrate that deterministic causal graph inference over black-box cipher output yields a universal framework for automated security margin analysis and cipher architecture identification. Using the .causal binary format for structured knowledge graph inference, we introduce CASI (Causal Amplification Security Index), a single-number metric that quantifies the structural weakness of reduced-round cipher output relative to the full-round baseline. Across six cipher families spanning four architectural classes — ARX stream (ChaCha20, Salsa20), SPN block (AES-128), ARX Feistel (Speck 32/64), Feistel with key-dependent S-boxes (Blowfish), and Feistel with fixed S-boxes (3DES) — CASI detection frontiers match published cryptanalysis results. On Speck 32/64, the CASI frontier aligns with Gohr's neural distinguisher (CRYPTO 2019) within one round, achieved without training, without GPU computation, and without cipher-specific tuning. Furthermore, .causal graph topology enables correct identification of all five architecture classes from reduced-round output alone. Key results: 6 ciphers × 28 round configurations analyzed; CASI frontier matches Gohr within 1 round on Speck 32/64; 100% blind architecture classification (5 classes); same 17 extraction strategies across all ciphers with no cipher-specific tuning. All code, cipher implementations, and generated .causal files are included in this deposit.

Demander à l'IA

Bookmark

View Full Paper