Work

BinXAuth: Compilation-robust binary authorship verification using dual-modality stylometric representations

Year: 2026
Type: article
Abstract: Binary authorship verification is essential for determining software provenance, conducting malware forensics, and investigating cyber-threats. Compiler transformations have been identified as a factor that can obscure persistent authorship evidence. Changes in compiler family, compiler version, and optimization level can alter instruction sequences and control-flow structures, leading to learned representations that capture toolchain-specific artifacts rather than author-related characteristics. In this study, we propose a compiler-confounding-aware dual-modality framework, BinXAuth, for binary authorship verification. BinXAuth jointly models normalized instruction sequences and instruction-enriched function-level control-flow graphs, treating compiler family, compiler version, and optimization level as nuisance domains during the representation learning process. The use of Transformer and graph isomorphism network encoders produces complementary sequential and structural representations that are fused into a shared stylometric space. These representations are optimized to maintain author-discriminative information while minimizing reliance on compilation-domain artifacts. We evaluated BinXAuth on a provenance-controlled, author-disjoint benchmark comprising 183 programmers, 1098 source programs, and 21,960 compiled binaries, spanning 20 GCC/Clang compilation domains. Under heterogeneous compilation, BinXAuth achieved an area under the receiver operating characteristic curve of 0.877 ± 0.004. This represents an improvement of 3.3 percentage points over the 0.844 ± 0.032 of VeriBin, the best-performing competing method evaluated in this study. Controlled cross-version and cross-compiler experiments, together with representation and objective ablations and domain-loss sensitivity analysis, further demonstrated the improved relative transfer of BinXAuth under compiler-induced distribution shifts. Our results indicate that BinXAuth achieves enhanced compilation robustness while preserving binary authorship verification performance across the evaluated compilation conditions. Here, compilation robustness refers to naturally occurring compiler-, version-, and optimization-induced shifts, and does not imply robustness to deliberate adversarial or anti-forensic transformations.
Language: English
Open Access status: hybrid
Estimated APC paid: $2,880
OpenAlex ID: W7233601217
FWCI:
Cites: 23
Cited by:
Related to: