Cross-Language Transfer Learning for Detecting Vulnerabilities in LLM-Generated Code
Global Media and Social Sciences Research Journal · 01.02.2026 · 7 (1) · 168-174
DOI: 10.71465/gmssrj161 ↗Kurz gefasst
Fachartikel in der Zeitschrift Global Media and Social Sciences Research Journal, online erschienen am 01.02.2026. Verfasst von Lucas Ribeiro und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-SA 4.0).
Abstract (Originalsprache)
We train a multilingual vulnerability detector capable of analyzing LLM-produced code across C, C++, Java, Python, and Rust. Using aligned intermediate representations that abstract language-specific syntax, the model learns security-relevant behavior patterns transferable between languages. Across 5 language benchmarks, the model achieves an average AUC of 0.91, outperforming monolingual detectors by 18–34%. It is especially effective at detecting logic bugs in authentication and authorization workflows. The study shows that transferable security representations are essential when LLMs generate polyglot code.
Zitieren
APA
Ribeiro, L., Carvalho, A., & Moreira, D. (2026). Cross-Language Transfer Learning for Detecting Vulnerabilities in LLM-Generated Code. Global Media and Social Sciences Research Journal, 7(1), 168-174. https://doi.org/10.71465/gmssrj161
Vancouver
Ribeiro L, Carvalho A, Moreira D. Cross-Language Transfer Learning for Detecting Vulnerabilities in LLM-Generated Code. Global Media and Social Sciences Research Journal. 2026;7(1):168-174. doi:10.71465/gmssrj161
BibTeX
@article{Ribeiro2026,
author = {Ribeiro, Lucas and Carvalho, Ana and Moreira, Daniel},
title = {Cross-Language Transfer Learning for Detecting Vulnerabilities in LLM-Generated Code},
journal = {Global Media and Social Sciences Research Journal},
year = {2026},
volume = {7},
number = {1},
pages = {168--174},
doi = {10.71465/gmssrj161}
}
EUVIMED