EUVIMEDEUVIMED entdecken
Crossref

Can You Trust the Vectors in Your Vector Database? Black-Hole Attack from Embedding Space Defects

Hanxi Li, Jianan Zhou, Jiale Lao, Yibo Wang, Zhengmao Ye, Yang Cao, Junfen Wang, Mingjie Tang

Proceedings of the VLDB Endowment · 01.09.2026 · 20 (1) · 31-45

DOI: 10.14778/3845598.3845601 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Proceedings of the VLDB Endowment, online erschienen am 06.11.2026. Verfasst von Hanxi Li und 7 weiteren Autor:innen. Die Arbeit nennt 79 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Vector databases serve as the retrieval backbone of modern AI applications, yet their security remains largely unexplored. We propose the Black-Hole Attack, a poisoning attack that injects a small number of malicious vectors near the geometric center of the stored vectors. These injected vectors attract queries like a black hole and frequently appear in the top- k retrieval results for most queries. This attack is enabled by a phenomenon we term centrality-driven hubness : in high-dimensional embedding spaces, vectors near the centroid become nearest neighbors of a disproportionately large number of other vectors, while this centroid region is nearly empty in practice. The attack shows that vectors in a vector database cannot be blindly trusted: geometric defects in high-dimensional embeddings make retrieval inherently vulnerable. Based on this insight, we propose four attack paths tailored to different attacker capabilities. Our experiments show that up to 94.4% of queries are successfully attacked. Additionally, we study two directions of defense: hubness mitigation and detection-based filtering. Hubness mitigation either significantly reduces retrieval accuracy or provides only limited protection, while the detection-based defense is effective against some attack paths but fails against others. A robust and adaptive defense thus remains an open problem, and our findings indicate that vector databases require more careful treatment of security.

Zitieren

APA

Li, H., Zhou, J., Lao, J., Wang, Y., Ye, Z., Cao, Y., Wang, J., & Tang, M. (2026). Can You Trust the Vectors in Your Vector Database? Black-Hole Attack from Embedding Space Defects. Proceedings of the VLDB Endowment, 20(1), 31-45. https://doi.org/10.14778/3845598.3845601

Vancouver

Li H, Zhou J, Lao J, Wang Y, Ye Z, Cao Y, et al. Can You Trust the Vectors in Your Vector Database? Black-Hole Attack from Embedding Space Defects. Proceedings of the VLDB Endowment. 2026;20(1):31-45. doi:10.14778/3845598.3845601

BibTeX
@article{Li2026,
  author = {Li, Hanxi and Zhou, Jianan and Lao, Jiale and Wang, Yibo and Ye, Zhengmao and Cao, Yang and Wang, Junfen and Tang, Mingjie},
  title = {Can You Trust the Vectors in Your Vector Database? Black-Hole Attack from Embedding Space Defects},
  journal = {Proceedings of the VLDB Endowment},
  year = {2026},
  volume = {20},
  number = {1},
  pages = {31--45},
  doi = {10.14778/3845598.3845601}
}
Kostenlos starten →