Self-replying bot networks have been a thing for some time now (2012), tested locally on forums. Why do you think they have such an unnaturally high number of followers?
The AI are pretty dumb, they follow a sort of 'Q&A' script - usually pre-scripted, with some limited flexibility. Basically, imagine a two chat bots talking.
Easy way to spot a self-replying bot network:
1) Same topic
2) Same/similar syntax (may change with upgrades in future)
3) Following/talking to each other
4) No 'life' outside of internal communications with one another
5) Leader-follower network/designation (one bots 'leads' the conversation, the other 'replies' - you'll never see a comment 'out of order' or 'spontaneous', but could be faked in modified or upgraded variants)
6) Same wake/sleep cycles (a bit strange an American and a Singaporean would be up and about at the same time of night, don't you think?)
7) Continuous posting.
8) Lack of meaningful content.
9) Lack of responsiveness when challenged (usually handler has to take over), or
10) Non-sensical replies
11) Grammatically unsound constructs that are unnatural and broken (like a chatbot).
USAF etc have the ability to register hundreds of accounts across proxies in seconds on twitter, and have them start following each other.
Ask yourself when they were created, compare creation dates, posting times. Real people naturally get tired, fall out of sync, work actual jobs.
I bet you Cicada3301 is the NSA's recruitment programme. Which is why Ches - NSA bot tool - was plugging it.
They decode 'alien' signals. Exolinguistics, I believe it is called.
Don't quote me on that. Not an agent.