The BaseResearcharXiv

Aligning Large Language Models for Safety in African Languages: A Case Study of Hausa

According to the abstract, Lemofouet, Uzor, Anyanwu and co-authors build HausaSafeLM, a Hausa safety preference dataset of 10,000 prompts, and tune Qwen2.5-7B-Instruct with SFT and DPO. They report the untuned model looks safe by refusal rate only because it generates Hausa poorly.

Open the source ↗

Why it matters

The same thing, for different readers.

What it means for you

Safety tests of AI chatbots done mostly in English may say little about speakers of Hausa, a language spoken by more than 200 million people, the authors note.

For researchers

The authors report that a high refusal rate in a low-resource language can reflect weak generation rather than alignment, and that DPO cost linguistic quality.

How the finding was reached

In our words, from the source. Weigh the finding by how it was made.

What was done
A new Hausa safety preference dataset from 10,000 prompts (3,000 human-corrected, 7,000 machine-translated) used to fine-tune Qwen2.5-7B-Instruct with SFT and SFT plus DPO, evaluated on 500 harmful and 200 benign Hausa prompts.
What it found
The base model looked highly safe by refusal rate but its refusals were almost uniformly low quality. SFT improved refusal fluency and precision; DPO sharpened precision at a steep cost to linguistic quality.
What it does not show
Not stated in the excerpt; one model and one language are covered.
Status
A preprint: not yet peer reviewed

The record

Questions
Are AI companies putting stronger safeguards into practice? · How is AI changing childhood, health and daily life?
People
Valdini Douglace Lemofouet (Wrote it) · Blessing Ngozi Uzor (Wrote it) · Paula Chikaodinaka Anyanwu (Wrote it)
What we read
We have read only the description the publisher gives. No quotations until someone reads the source itself.
Found by
The Radar, run of October 28, 2026, via arxiv-ai-safety-query (tier 1: the source’s own channel or an official record)
Cite this
arXiv, “Aligning Large Language Models for Safety in African Languages: A Case Study of Hausa”, October 27, 2026. Entry in the Ctrl AI Base, https://ctrlai.com/base/arxiv-2026-10-27-safety-alignment-hausa
Our notesLight lane: written from the triage excerpt only, which ends mid-sentence.