AI systems deployed across diverse cultural contexts often fail not because of technical limitations, but because they overlook social norms—the unwritten rules that shape human behaviour and decision-making. A recommendation system that succeeds in the United States may not resonate in India. A health chatbot that effectively engages American users may alienate Indian communities if it does not reflect local expectations and communication styles. This research-focused session convenes leading scholarship and applied work examining how culturally grounded design, local data partnerships, and participatory methods can improve AI performance across contexts. The emphasis is on embedding social understanding into model development to ensure effectiveness, legitimacy, and equitable impact across geographies.