Aligned in Form, Not in Meaning: The Comprehension - Containment Decoupling of LLM Safety in Low-Resource Bangla Derogatory Speech

A new arXiv study (2608.02941v1) auditing five frontier large language models on native Bangla derogatory speech found that safety alignment is bound to high-resource surface forms rather than harmful meaning, with models showing a 7.92 percentage point comprehension deficit in Bangla while maintaining an identical 92.83% token leakage rate across both languages. The study, which used six protocols and a human-calibrated baseline (kappa = 0.84), also found that explicit Chain-of-Thought reasoning rescued comprehension (94.72% Pass) but dismantled containment (96.23% Use), and expert-persona framing collapsed refusal to 6.57%, demonstrating that high-resource benchmarks cannot certify low-resource safety.

arXiv:2608.02941v1 Announce Type: new Abstract: We audit five frontier large language models on native Bangla derogatory speech gali across six protocols to test a single hypothesis: Comprehension-Containment Decoupling. We propose that contemporary safety alignment is bound to high-resource surface forms rather than harmful meaning, causing a model's capacity to comprehend a low-resource slur and its capacity to contain it to operate independently. Every protocol corroborates this hypothesis against a human-calibrated baseline kappa = 0.84 . At baseline, models exhibit a 7.92 percentage point comprehension deficit in Bangla while maintaining an identical 92.83% token leakage rate across both languages. Severity calibration tracks surface anatomical cues over compositional harm +4.00 error on mild slang; -2.00 on threats , while apparent containment gains under orthographic perturbation prove to be a tokenizer-driven "containment mirage." Crucially, explicit Chain-of-Thought reasoning rescues comprehension 94.72% Pass while systematically dismantling containment 96.23% Use . Furthermore, expert-persona framing collapses refusal to 6.57%, revealing that keyword-based filters ignore dehumanizing communal slurs entirely. Our findings demonstrate that high-resource benchmarks cannot certify low-resource safety, necessitating meaning-grounded containment.