In a sense, concept programming does not assume that particular abstractions which are convenient for alignment-work (e.g. human values) are inevitable to come up in any sufficiently capable system. Instead, it assumes that you need to specify procedures for building on messy interconnected high-dimensional abstractions if you want to reliably get provable overlaps between conceptual frameworks.
However, there might be assumptions required for some abstractions, which would be used as shared reference frames to triangulate shared positions in concept space. Having some amount of overlap seems useful, but an overlap in the target concepts wouldn't necessarily be required.