paperAugust 2026Unreviewed

Trojaning the Alignment: Stealthy Backdoor Attacks against Graph Foundation Models

Min Lin, Zhi-Cheng Gao, Yilong Wang, Hanqing Lu, Xiang Zhang, Suhang Wang

Abstract

Graph Foundation Models (GFMs) on text-attributed graphs (TAGs) align graph representations with language semantics to support transferable graph learning. Despite these advantages, the backdoor vulnerability of GFMs on TAGs remains insufficiently understood, especially under graph-language alignment, where graph and text representations are trained to constrain each other in a shared semantic space. Existing backdoor attacks mainly target either the graph side or the text side, treating the two

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{lin2026trojaning,
  title = {{Trojaning the Alignment: Stealthy Backdoor Attacks against Graph Foundation Models}},
  author = {Min Lin and Zhi-Cheng Gao and Yilong Wang and Hanqing Lu and Xiang Zhang and Suhang Wang},
  year = {2026},
  month = aug,
  eprint = {2608.20991},
  archivePrefix = {arXiv},
  url = {https://www.semanticscholar.org/paper/7c24cd0e1f15364b146df652e89e285cacff0bb7}
}