RESEARCHInvestigateNEXT 12 MONTHS
Language Shapes Instruction Hierarchy Compliance in Multilingual LLMs
arXiv cs.CL — Computation and Language
Factual evidence
What the source reports
Research introduces XIH-Bench, a new benchmark to evaluate instruction hierarchy compliance in multilingual LLMs, finding language impacts adherence.
Open sourceOneBench interpretation
Institutional assessment
So what
Instruction hierarchy compliance variations across languages impact global deployment safety and control for G-SIBs.
Do what
Brief your model risk and responsible AI teams on multilingual instruction hierarchy vulnerabilities.