paperMarch 2026Unreviewed

Detecting Data Poisoning in Code Generation LLMs via Black-Box, Vulnerability-Oriented Scanning

Shenao Yan, Shimaa Ahmed, Shan Jin, Sunpreet S. Arora, Yiwei Cai, Yizhen Wang, Yuan Hong

Abstract

Code generation large language models (LLMs) are increasingly integrated into modern software development workflows. Recent work has shown that these models are vulnerable to backdoor and poisoning attacks that induce the generation of insecure code, yet effective defenses remain limited. Existing scanning approaches rely on token-level generation consistency to invert attack targets, which is ineffective for source code where identical semantics can appear in diverse syntactic forms. We present

Framework mappings

OWASP Top 10 for LLM Applications

LLM04Data and Model Poisoning

MITRE ATLAS

AML.T0020Poison Training Data

Suggested from the entry's categories.

Cite

@misc{yan2026detecting,
  title = {{Detecting Data Poisoning in Code Generation LLMs via Black-Box, Vulnerability-Oriented Scanning}},
  author = {Shenao Yan and Shimaa Ahmed and Shan Jin and Sunpreet S. Arora and Yiwei Cai and Yizhen Wang and Yuan Hong},
  year = {2026},
  month = mar,
  eprint = {2603.17174},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2603.17174}
}