We also didn't have AI models that politicians believed could detect bad behavior on a mass scale. Implementing Stasi level mass surveillance would be very expensive back then, even if there was full access to all communication. Now the proposal is to make a model try to categorize your messages and flag you.

I believe that politicians believe that AI models can do this well without negative consequences. But I also think they forget that a model with 99.99% specificity applied to ten million messages will still falsely label 1000 as harmful.