We show that any problem that has a classical zero-knowledge protocol against the honest verifier also has, under a reasonable condition, a classical zero-knowledge protocol which is secure against all classical and quantum polynomial time verifiers, even cheating ones. Here we refer to the generalized notion of zero-knowledge with classical and quantum auxiliary inputs respectively.
Our condition on the original protocol is that, for positive instances of the problem, the simulated message transcript should be quantum computationally indistinguishable from the actual message transcript. This is a natural strengthening of the notion of honest verifier computational zero-knowledge, and includes in particular, the complexity class of honest verifier statistical zero-knowledge. Our result answers an open question of Watrous [Wat06], and generalizes classical results by Goldreich, Sahai and Vadhan [GSV98], and Vadhan [Vad06] who showed that honest verifier statistical, respectively computational, zero knowledge is equal to general statistical, respectively computational, zero knowledge.