mlsanityData to 14 Oct 2026

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

arXiv 2608.02941 ↗First posted 3 Aug 2026cs.CLcited by 0

On the map: LLM safety and security → Safety alignment, refusal and guardrails

Abstract (arXiv)

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.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Safety alignment, refusal and guardrails, one of 12 directions in LLM safety and security. Builds on 11 of its 20 references that are in the map. First posted on arXiv 3 Aug 2026.

Builds on

11 of 20 references.

Closest ancestors · 1

Same direction, newest first

  1. 2025The State of Multilingual LLM Safety Research: From Measuring the Language Gap to Mitigating It30 May

Other references · 10

  1. 2024LLMs instead of Human Judges? A Large Scale Empirical Study across 20 NLP Evaluation Taskscited by 243
  2. 2024Annotation alignment: Comparing LLM and human annotations of conversational safetycited by 23
  3. 2019Towards Robust Toxic Content Classificationcited by 22
  4. 2024Dialectal Toxicity Detection: Evaluating LLM-as-a-Judge Consistency Across Language Varietiescited by 12
  5. 2025Does Chain-of-Thought Reasoning Really Reduce Harmfulness from Jailbreaking?cited by 9
  6. 2025No for Some, Yes for Others: Persona Prompts and Other Sources of False Refusal in Language Modelscited by 9
  7. 2026Expert Personas Improve LLM Alignment but Damage Accuracy: Bootstrapping Intent-Based Persona Routing with PRISMcited by 9
  8. 2025Chain-of-Lure: A Universal Jailbreak Attack Framework using Unconstrained Synthetic Narrativescited by 5
  9. 2024Cross-lingual Offensive Language Detection: A Systematic Review of Datasets, Transfer Approaches and Challengescited by 2
  10. 2025All You Need is "Leet": Evading Hate-speech Detection AIcited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026The Illusion of Cross-Lingual Safety in Low-Resource Languages
  2. 2026Low-Resource Safety Failures Are Action Failures, Not Representation Failures
  3. 2026Sycophancy as a Multilingual Alignment Failure: How Safety Degrades Across Languages, Topics, and Models
  4. 2025Phonetic Perturbations Reveal Tokenizer-Rooted Safety Gaps in LLMs
  5. 2026The Wrong Kind of Right: Quantifying and Localizing Misfired Alignment in LLMs
  6. 2025The Alignment Veto: How Safety Training Suppresses Cultural Knowledge in LLMs
  7. 2026SHARD: Safe and Helpful Alignment via Self-Reframing Distillation
  8. 2026Alignment Backfire: Language-Dependent Reversal of Safety Interventions Across 16 Languages in LLM Multi-Agent Systems