ABench: Adversarily Benchmarking Large Language Models across Taxonomized Values.

arXiv:2610.27611v1 Announce Type: cross Abstract: Ensuring the alignment of Large Language Models (LLMs) with diverse cultural values is critical for their safe global deployment. However, existing alignment benchmarks are predominantly English-centric or conf