AI safety is designed in the West, and failing users everywhere

OpenAI’s recent decision to pause the release of a voice feature highlights gaps in safety testing for non-Western languages and cultural contexts. As AI development moves rapidly, current safeguards often fail to account for the specific harms experienced by users outside of English-speaking regions. This oversight suggests that existing safety frameworks may be insufficient for a global user base.