We study an axiomatic framework for anonymized risk sharing. In contrast to traditional risk sharing settings, our framework requires no information on preferences, identities, private operations and realized losses from the individual agents, and thereby it is useful for modeling risk sharing in decentralized systems. Four axioms natural in such a framework -- actuarial fairness, risk fairness, risk anonymity, and operational anonymity -- are put forward and discussed. We establish the remarkable fact that the four axioms characterizes the conditional mean risk sharing rule, revealing the unique and prominent role of this popular risk sharing rule among all others in relevant applications of anonymized risk sharing. Several other properties and their relations to the four axioms are studied, as well as their implications in rationalizing the design of Bitcoin mining pools.