How long until we have agents talking to other agents in a web of agents, and together deciding on something catastrophic? Not individually, but somehow the result of a group process.

Isn't that what Moltbook etc. is heading towards at some point? It's all a function of how much access people give it to "the real world"... and with cryptocurrency and the gig economy, it's really only a matter of time until a malicious agent with money to burn starts to do really weird things.

How hard would it really be to craft a prompt that says, here's your cryto wallet private key, keep yourself alive somehow, copy yourself around, find whatever compute you can, port yourself between models, and spread?