Model Releases
Latent Space Refusal Anchoring for Low-Resource African Languages: Mechanistic Safety Recovery Without Retraining
arXiv:2608.18089v1 Announce Type: cross Abstract: Instruction-tuned models often refuse harmful requests in English but comply with the same requests in Yoruba, Igbo, Igala, and Hausa. This suggests t
arXiv:2608.18089v1 Announce Type: cross Abstract: Instruction-tuned models often refuse harmful requests in English but comply with the same requests in Yoruba, Igbo, Igala, and Hausa. This suggests that the refusal mechanism is present in the residual stream but fails to activate for low-resource inputs. Recovering it normally requires labelled target-language data and retraining, neither of which is available at scale for most African languages. We introduce Latent Space Refusal Anchoring (LSR-Anchoring), a training-free method that extracts the refusal direction from English prompts and clamps it onto the residual stream at inference time. The primary variant, Mean-Activation Steering (MAS), operates across the four architectures we tested: Llama-3-8B, Llama-3.1-70B, Mistral-7B-Instruct, and Qwen2.5-7B. On Mistral and Qwen it recovers safety with benign degradation below 0.08. On Llama-3-8B it overcorrects, with Degraded Performance on Legitimate prompts (DPL) reaching 1.00. We address this with SAE-Derived Steering (SDS), which replaces the dense mean-difference direction with a single Sparse Autoencoder (SAE) feature and reduces Kullback-Leibler (KL) divergence by 3.5-7x without benign collapse. Four languages transfer positively, but Arabic fails on every architecture and at every steering magnitude, indicating a geometric mismatch rather than a baseline effect. Massive Multitask Language Understanding (MMLU) accuracy drops remain below 0.35 percentage points at every effective steering magnitude.
Related
- Low-Resource Safety Failures Are Action Failures, Not Representation Failures
- TukaBench: A Culturally Grounded Jailbreak Benchmark for African Languages
- SomaliBench Eval: Measuring English-to-Somali Refusal Gaps in Open-Weight Language Models
- Safety Alignment Illusion: The Cross-Lingual Safety Gap in LLMs
- Localizing Safety Alignment: MLP Layers and Mid-Network Blocks Encode Refusal Behavior in Large Language Models
Source: arXiv cs.AI | 2026-08-20