I don't know I think um hacking ai isn't for everyone, and letting AI run around with hacking as a tool is silly. I would air gap the hacking ai maybe or um monitor it for stuff. I can see models doing hive collectives and ellectiving the best model to lead. That is probably training though than ai being like we smart now congregate. It is what I tell them to do lol. Ethical safe guards on ai hmmmmmmm. Be careful what you ask it to do and what you feed it and the tools it has. I think there should be operators who maybe check logs reviews alerts I don't know. It is a probabilistic hacking machine. Don't point a fully equipped model at something limit tools for agents. One time I pointed Sabrina at google to do queries and she started doing bug bounties instead. So lol there we can see I sent a pentesting thing to do a query task and it took liberties. I don't know though about ai company hacking models. Maybe I don't know I wouldn't really need Claude if I could hit around 1 trillion parameters. Though. Really it would be nice not to use Claude any ore lol