Or at a higher level, at the ISP level.

Targeted via DNS tunneling and all.

fudge the up/down votes to make it look like it's been seen but not reacted to.

but do you need to burn cycles on AI to keep these people engaged? if someone is spamming stuff you don't want seen have them throw out a basic response and then shadowban or just straight-up ban them. if they're very negative bad actor types just give em the boot

Enough frustrated people will use AI to quickly generate the code for an alternative platform to avoid this bubble system.

It will be individual platforms all the way down...oh wait.

On platforms like Facebook or YouTube where the feed is algorithmically generated and you can't easily view a filtered list of topics (like Reddit) something like this would be very easy.

The interactions don't even need to be generated by AI, it just needs to keep you seeing interactions with other people in your social status circle. And if you try to venture too far outside of that it shadow bans you.

Heck I'd be surprised if the way the news feed algorithms work today they don't already do something like this, as a byproduct of optimising for viewership.

They'd just need to take it a bit further by preventing you from seeing viewpoints outside your circle. So taking the WWII example, people in the Nazi group would not be able to see pro-Ally content. All they'd see about Allies would be content that paints them in a bad light, and vice versa.