Security Knowledge Distillation for Enhancing LLM Safety in Code Generation
American Journal of Electrical Engineering and Technology · 31.01.2026 · 7 (1) · 88-94
DOI: 10.71465/ajeet3630 ↗Kurz gefasst
Fachartikel in der Zeitschrift American Journal of Electrical Engineering and Technology, online erschienen am 31.01.2026. Verfasst von Daniel Thompson und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
We propose a distillation framework that transfers secure coding expertise into a compact model trained using expert demonstrations, static-analysis labels, and secure patterns. The distilled model (7B parameters) achieves 83% security compliance, nearly matching GPT-4’s 89% performance while requiring significantly less compute. Evaluated across CWE-20, CWE-78, CWE-89, the distilled model reduces injection vulnerabilities by 62% relative to baseline models. This work demonstrates that targeted distillation can teach smaller models to respect security best practices.
Zitieren
APA
Thompson, D., Mitchell, S., & Li, J. (2026). Security Knowledge Distillation for Enhancing LLM Safety in Code Generation. American Journal of Electrical Engineering and Technology, 7(1), 88-94. https://doi.org/10.71465/ajeet3630
Vancouver
Thompson D, Mitchell S, Li J. Security Knowledge Distillation for Enhancing LLM Safety in Code Generation. American Journal of Electrical Engineering and Technology. 2026;7(1):88-94. doi:10.71465/ajeet3630
BibTeX
@article{Thompson2026,
author = {Thompson, Daniel and Mitchell, Sarah and Li, Jason},
title = {Security Knowledge Distillation for Enhancing LLM Safety in Code Generation},
journal = {American Journal of Electrical Engineering and Technology},
year = {2026},
volume = {7},
number = {1},
pages = {88--94},
doi = {10.71465/ajeet3630}
}
EUVIMED