When Safety Speaks a Language: A Mechanistic Analysis of Safety-Language Identity Entanglement in LLMs
Safety alignment of large language models (LLMs) degrades across languages, yet the internal mechanism driving this asymmetry remains poorly understood. Our work, therefore, presents a systematic mechanistic analysis of multilingual safety using sparse autoencoder (SAE) features, sparse interpretable directions in the residual stream associated with harmful and harmless model behavior across three instruction-tuned LLMs, eight languages, and all model layers. We observe that safety-relevant features are architecture-dependent in terms of where they are located and how they are distributed across layers. Additionally, they are geometrically entangled with language identity and exhibit cros...
| Authors | Apoorva Upadhyaya, Sandipan Sikdar |
|---|---|
| Primary category | cs.CL |
| Published | 30 August 2026 |
| Last updated at source | 30 August 2026 |
Silytics shows public metadata from arXiv and links to the original. Information may be incomplete or out of date. Last refreshed 17 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.