A basic requirement for a cryptographic key distribution protocol (of any kind, classical or quantum) is that the generated key can be used as if it was a perfect key, i.e., a uniform random value unknown to any adversary. As reported recently (quant-ph/0512021), for quantum key distribution (QKD), this property is not implied by most of the established security definitions and security proofs. In particular, one-time pad encryption might be insecure if one uses a key generated by a QKD protocol. In the talk, we discuss possible solutions to this problem.