paperAugust 2026UnreviewedOpen access

Pervasive Backdoor Vulnerabilities in Genomic Foundation Models

Shiwen Ni, Qianning Wang, Chi Wei, Xiaomin Ni, Shuai-Min Li, Zixin Zhao, Hui Li, Rongrong Ji, Teng Wang, Min Yang

bioRxiv

Abstract

Genomic foundation models are increasingly used to interpret and design DNA sequences, yet their susceptibility to training-data manipulation remains poorly understood. Here we systematically evaluate backdoor poisoning across three model families, seven parameter scales ranging from 50 million to 7 billion, and 18 genomic classification tasks. We introduce two complementary 48-nucleotide triggers: a composition-matched synthetic sequence and a biologically grounded trigger derived from transpos

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

@article{ni2026pervasive,
  title = {{Pervasive Backdoor Vulnerabilities in Genomic Foundation Models}},
  author = {Shiwen Ni and Qianning Wang and Chi Wei and Xiaomin Ni and Shuai-Min Li and Zixin Zhao and Hui Li and Rongrong Ji and Teng Wang and Min Yang},
  year = {2026},
  month = aug,
  journal = {bioRxiv},
  doi = {10.64898/2026.07.30.741642},
  url = {https://www.semanticscholar.org/paper/9a304471e350bc4004450beaf57b8eead3676e2b}
}