As an offline-platform can make all sorts of bureacratic rules designed to censor those they don't like, this problem is far-far worse on a software-driven plat

As an offline-platform can make all sorts of bureacratic rules designed to censor those they don't like, this problem is far-far worse on a software-driven platform. Imagine facebook wants to censor anti-Biden, or anti-Trump, posts. They needn't design an algorithm that does explicitly that - that could be scrutinised under review. Instead: design an 'anti-spam' algorithm. Within this anti-spam algorithm, characteristic Y is flagged. Characteristic Y just happens to be heavily-correlated to anti-biden/anti-trump posts. A bit of filtering later (for 'anti-spam measures' of-course), and you have achieved your censorship without being too overt.