This article discusses the provable security of block-cipher-based hash functions. It introduces a new model called a weak ideal cipher model. In this model, an adversary is allowed to make key-disclosure queries to the oracle as well as encryption and decryption queries. A key-disclosure query is a pair of a plaintext and a ciphertext, and the reply is a corresponding key. Thus, in this model, a block cipher is random but completely insecure as a block cipher. It is shown that collision resistant hash functions can be constructed even in this weak model.