An actual sensible legal regime could be a bit like the CDA safe harbor, or at least more like what CDA safe harbor should be. If an intermediary like Shopify provides services to its clients, does not help users choose which of its clients to work with, declines to exercise discretion as to which clients to work with, and cooperates with court orders, then it should be immune to claims that it chose the wrong clients.
I agree with others in this thread - they should provide a layer of actual humans to help shop owners. And have a process for dealing with AI/ML gone rogue - like the address verification system in this example - where a set of humans can override and make sure a shop owner can continue to operate. Heck send someone to PR to help validate addresses every month.