I don't think that offers meaningful protection. A malicious government could acquire or produce real CSAM, or at least a convincing fake of it, then modify that image subtly to match the hash of the political content they were targeting (I believe such engineered collisions have already been demonstrated.) They would then be able to get that manipulated CSAM into as many independent CSAM databases as they wanted, simply by uploading it to the web then reporting it to all those independent databases.