Unless there is end-to-end encryption (which gmail does not support, partially because US government requires Google to monitor accounts supposedly belonging to

Unless there is end-to-end encryption (which gmail does not support, partially because US government requires Google to monitor accounts supposedly belonging to terrorists etc), the server always "knows" your email in plain text. It seems like processing user email text through a program that actively "works" on the text (v.s. simply encrypting/decrypting/transmitting the text) is generally not considered a privacy concern. I feel like this is a bit tricky given the evolution in ML/AI. By making "feeding user email into any program" acceptable, the chance that rule-breaking incidents (e.g., using email text to train models) is unnecessarily increased imo.