From Symmetry toward Weak Asymmetry: Secure Steganography under the Receiver's Partial Channel Knowledge

Provably secure steganography enables covert communication by disguising secret messages as normal communication content while guaranteeing computational indistinguishability. Recent work typically assumes that the sender can construct stegotext based on explicit conditional distributions, while the receiver either has access to the same distributions for decoding or has no such access at all. In many realistic communication settings, however, the receiver's access lies between these two extremes, a regime that is not captured by either assumption. We therefore introduce and study a new setting, called weakly asymmetric steganography, in which the sender has access to the full conditional distributions used for stegotext construction, while the receiver only has partial knowledge derived from them. We design provably secure steganography schemes under several representative forms of receiver-side knowledge. Specifically, we consider binary alphabets in which the receiver only knows whether a symbol's probability exceeds a given threshold, as well as general alphabets where the receiver only observes a possibly partial ranking of symbols by probability. We evaluate our rank-based constructions in language-model-based text steganography under both limited receiver-side knowledge and precision-induced distribution mismatch. The results show that weakly asymmetric steganography can achieve non-trivial embedding capacity, and further suggest the potential of this setting in distribution-mismatch scenarios.