让大模型在低资源语言中也更安全，通过锁定语义瓶颈实现跨语言安全对齐。
LASA: Language-Agnostic Semantic Alignment at the Semantic Bottleneck for LLM Safety
- 在模型语义瓶颈层直接对齐安全属性，避开语言表面差异。
- 多语言测试中攻击成功率从24.7%降至2.8%，低资源语言表现显著提升。
- 适合关注大模型跨语言安全性与可解释性对齐的研究者。
大型语言模型（LLMs）在高资源语言中通常表现出良好的安全性，但在低资源语言中却存在严重漏洞。我们归因于模型的语义理解能力与偏向高资源语言的安全对齐之间存在不匹配。实证发现，模型中间层存在语义瓶颈——该层表示几何主要由共享语义内容决定，而非语言身份。基于此，我们提出语言无关语义对齐（LASA），将安全对齐直接锚定在语义瓶颈层。实验表明，LASA显著提升所有语言的安全性：在LLaMA-3.1-8B-Instruct上，平均攻击成功率为2.8%（原为24.7%），在Qwen2.5和Qwen3 Instruct（7B-32B）模型上维持在3%-4%。分析与方法揭示了模型安全性应扎根于语言无关的语义空间，而非表层文本。
原文摘要 · Abstract (English)
Large language models (LLMs) often demonstrate strong safety performance in high-resource languages, yet exhibit severe vulnerabilities when queried in low-resource languages. We attribute this gap to a mismatch between language-agnostic semantic understanding ability and language-dominant safety alignment biased toward high-resource languages. Consistent with this hypothesis, we empirically identify the semantic bottleneck in LLMs, an intermediate layer in which the geometry of model representations is governed primarily by shared semantic content rather than language identity. Building on this observation, we propose Language-Agnostic Semantic Alignment (LASA), which anchors safety alignment directly in semantic bottlenecks. Experiments show that LASA substantially improves safety across all languages: average attack success rate (ASR) drops from 24.7% to 2.8% on LLaMA-3.1-8B-Instruct and remains around 3-4% across Qwen2.5 and Qwen3 Instruct models (7B-32B). Together, our analysis and method offer a representation-level perspective on LLM safety, suggesting that safety alignment requires anchoring safety understanding not in surface text, but in the model's language-agnostic semantic space.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。