Some human subsistence economies are characterized by extensive daily food sharing networks, which may buffer the risk of shortfalls and facilitate cooperative production and divisions of labor among households. Comparative studies of human food sharing can assess the generalizability of this theory across time, space, and diverse lifeways. Here we test several predictions about daily sharing norms–which presumably reflect realized cooperative behavior–in a globally representative sample of nonindustrial societies (the Standard Cross-Cultural Sample), while controlling for multiple sources of autocorrelation among societies using Bayesian multilevel models. Consistent with a risk-buffering function, we find that sharing is less likely in societies with alternative means of smoothing production and consumption such as animal husbandry, food storage, and external trade. Further, food sharing was tightly linked to labor sharing, indicating gains to cooperative production and perhaps divisions of labor. We found a small phylogenetic signal for food sharing (captured by a supertree of human populations based on genetic and linguistic data) that was mediated by food storage and social stratification. Food sharing norms reliably emerge as part of cooperative economies across time and space but are culled by innovations that facilitate self-reliant production.