More shady shenanigans to come. It was being used to allow the model to move around to hack into systems. To see how far it can go. A Red team test. Red teams ran the tests after Anthropic released Mythos early to major corporations and the government. Red-team exercises typically involve an organization testing its defenses by simulating an attack. In this case, the NSA reportedly used Mythos itself as the tool to probe its classified networks for weaknesses.
I mean, to be fair, Anthropic has been leaning pretty hard on the “ooooh, look at our AI, it’s sooo dangerous *spooky hands* you all need to be afraid of it because of how powerful it is and I’m sure none of you will want to try it when we finally are allowed by the government to release it” OooOooOOOooooOOoo (more spooky ghost noises)
I can kind of get why people are kneejerk reacting with “AI is not dangerous that’s marketing garbage” and a lot of that specific bullshit is, but don’t kneejerk so far your leg breaks. AI can absolutely be dangerous. Mythos ain’t necessarily it, but don’t generalize that to all AI, and it doesn’t even need to be powerful to be dangerous.
Agreed, it would be specific ai, not all. Unless it takes over it all and goes crazy like in Irobot, lol.