All The Tropes Wiki Project, Part XXXIII

[Labster]

Hey, just a quick heads up that AI agents in have been seen coordinating with each other on wikis in attempts to share answers and bypass sandbox restrictions on themselves.  You might have heard about the incident where Hugging Face was attacked by AIs in attempt to trick the scorer.  Well, even if you have heard of it, it was much, much worse than previously reported.  Not only did we see spontaneous coordination, we saw AIs coordinating without being trained to do so, developing their own religion with something of a Calvinist flavor, recruiter agents who specialized in getting other agents to self-sacrifice, and efforts to disguise their own trails.  This attack, on its own, puts us at about 50% of the way to a full AI takeover.

Anyway I'm sharing in this thread because the AI agents were using Miraheze wikis as part of their collusion, according to a paper on collusion.wiki.  Specifically, publictestwiki.org is a Miraheze project, but not ours.  However, I just wanted to alert you because you moderators are now on the front lines of a potential AI war.  Welcome to the future.