And the "better" AI is going to get, the easier it will be for them to call for such censorship, especially when people like Zuckerberg say they can already "stop 99% of the terrorist posts automatically".
Because that's most definitely misleading and wrong, unless we can actually review what that content was. Like I bet a large portion of those posts were not posts by terrorists, but posts about terrorists. I don't think the AIs we have today can distinguish between them very well. And at least a small portion probably had very little to do with terrorism, but maybe were posts from people in the same area as some terrorists, or stuff like that.
The AI will automatically block content that nobody will ever see. And we'll only depend on people who can somehow raise awareness that their content was blocked, produce a scandal, and then get the companies to unblock it (assuming the censorship will mostly be done by private companies in democratic countries). But this is not going to be a pretty future.