Russell and Wang [22] recently introduced an elegant, information-theoretic notion called entropic security of encryption: they required that the cipher text leak no predicate of the plaintext (similar to semantic security [10]) but only as long as the distribution on messages has high entropy from the adversary’s point of view. They show that this notion of security can be achieved with very short keys for entropically rich message spaces. Canetti et al [6, 7] had previously constructed hash functions which satisfy a similar entropic security condition. The output of such hash function leaks no partial information about the input, provided the input has sufficiently high entropy. This paper studies entropic security in general, and its application to the encryption of high-entropy messages. • We elucidate the notion of entropic security. Our results apply to all entropically-secure primitives, including both encryption and hash functions. We strengthen the formulation of [6, 7, 2...