Systematic Review of LLM Safety Alignment in Low-Resource Languages
August 18, 2026
This review of 50 studies identifies a significant gap in safety guarantees for low-resource and multilingual settings. It finds that translated English benchmarks fail to capture culturally specific harms and categorizes alignment methods into data, objective, and mechanistic approaches.
HOW THIS AFFECTS YOU
●
researcherYou can use this taxonomy to identify gaps in cross-lingual transferability and safety optimization.
●
policyYou should be aware that safety alignment for non-English languages remains a critical and unaddressed vulnerability.