Benchmark RadarTukaBench
TUKABENCH is a benchmark for jailbreak evaluation in seven African languages, extending JailbreakBench with human translations, cultural adaptations, and code-switched prompts. It assesses LLM safety in low-resource languages using metrics like Refused, Jailbroken, and Deflection, with human validation of LLM-as-a-judge.
- Released
- 2026-05-31
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
The benchmark addresses a gap in safety evaluation for low-resource African languages, showing that models are more vulnerable to jailbreak prompts in these languages. It could inform safer deployment of LLMs in multilingual contexts.
Motivation
Safety evaluation of Large Language Models (LLMs) remains heavily English-centric, leaving Low-Resource Languages (LRLs), particularly African ones, critically underexplored.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.