One of the ways in which AI produces and reproduces power and politics is within datasets that are partial and biased. Data is a political tool which is always subjective and never whole. After its collection, the data is often disconnected from its history, people, and context of collection, which becomes an instance of power and politics of data as it masks its underlying subjective meanings and serves the purpose of objective quantification. Specialized data, consisting of information pertaining to only a certain group of people and in certain specific contexts and scenarios, is reused in generalized ways. Qualitative aspects of data are generalized and quantified, thus losing the essence of information it contains. Because of the loss of context, the data remains partial and creates problems, like: (1) people lose control over their data and how its continuous analysis impacts them; (2) algorithms built on the data produce results that are stereotypically biased; (3) and reproduce power relations. In the Southern context, especially in India caste is an integral part of demographic data collected from people. “General” becomes the category masked with privilege, while “Schedule Caste,” “Schedule Tribe,” and “Other Backward Classes” overexposes the groups within. The data collected around this framework inherently divides the people based on whether they are “general” or not, and thus teaches the AI trained on it to do the same. This chapter closely examines the meta narratives conveyed by different databases that have a table consisting of a column titled “caste” and the suggested machine learning models for them on famous database community Kaggle.com ( https://www.kaggle.com/search?q=caste ) to argue that the upcoming generations of AI systems in India are set to digitally repurpose the race bias of the Global North into caste bias and perpetuate digital inequality.