Hard Negatives Reveal What Easy Negatives Hide: Cross-Lingual Harmfulness Representations Degrade with Resource Tier Under Hard Negatives

Excerpt

arXiv:2609.27758v1 Announce Type: new Abstract: Safety alignment in large language models is trained primarily in English, and recent work reports that the underlying harmfulness representation survives translation: English-trained probes separate harmful from harmless prompts almost as well in low-resource languages as in English. This has been taken as evidence that cross-lingual refusal failures mainly reflect calibration rather than representation quality. We show that this conclusion depend