EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Mathematical Foundations of Cross-Lingual Vulnerabilities in LLMs: Latent Space Entanglement and Token Fragmentation

Umar Hasan, Muhammad Ali Nayeem

Mathematics · 26.05.2026 · 14 (11) · 1849

DOI: 10.3390/math14111849 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Mathematics, online erschienen am 26.05.2026. Verfasst von Umar Hasan und 1 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 64 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Qassim University.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Large Language Models (LLMs) are typically safety-aligned using high-resource language data, but it remains unclear whether these constraints transfer reliably across distinct linguistic manifolds. This study examines the mathematical foundations of cross-lingual guardrail degradation using Bengali as a low-resource test case. We evaluate Meta-Llama-3-8B, Gemma-2-9B, and Llama-Guard-3 through an automated English-to-Bengali translation pipeline, paired statistical testing, latent-space visualization, and tokenization-based structural analysis. The results show a statistically significant increase in Gemma-2’s Attack Success Rate from 32.0% in English to 41.2% in Bengali (p

Förderung

  • Qassim University

Zitieren

APA

Hasan, U., & Nayeem, M. A. (2026). Mathematical Foundations of Cross-Lingual Vulnerabilities in LLMs: Latent Space Entanglement and Token Fragmentation. Mathematics, 14(11), 1849. https://doi.org/10.3390/math14111849

Vancouver

Hasan U, Nayeem MA. Mathematical Foundations of Cross-Lingual Vulnerabilities in LLMs: Latent Space Entanglement and Token Fragmentation. Mathematics. 2026;14(11):1849. doi:10.3390/math14111849

BibTeX
@article{Hasan2026,
  author = {Hasan, Umar and Nayeem, Muhammad Ali},
  title = {Mathematical Foundations of Cross-Lingual Vulnerabilities in LLMs: Latent Space Entanglement and Token Fragmentation},
  journal = {Mathematics},
  year = {2026},
  volume = {14},
  number = {11},
  pages = {1849},
  doi = {10.3390/math14111849}
}
Kostenlos starten →