A plethora of known methods describe how policy-breaking transmissions can be conducted, e.g., through network censorship circumvention channels or through local covert channels. While some of these channels have very low capacities, only a few studies have been published on improving the content-compressibility for such channels. However, existing methods are either not useful for compressing small data, require to crawl third-party content (e.g., websites), or can only transfer data with small bitrates. Nevertheless, in reality typical payloads are often small (e.g., user credentials, status updates, or chat messages) and usually remain uncompressed, leaving more traces for an adversary to analyze. We introduce AMPhitryon, a novel approach to compress small payloads \(<\) 100 B (especially 4–20 B). It can be used with existing tools to compress small payloads, while larger payloads can be compressed with arbitrary compressors. AMPhitryon employs a local dictionary that is compiled incrementally in a context-specific setting (e.g., the context could be a network flow or a chat). Our evaluation shows that AMPhitryon achieves higher compression ratios for short messages than existing techniques. It reaches the lowest (best) compressibility ratios (down to 0.30) with the smallest payloads (<20 B).