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.
AI can be used hack systems! I am not saying it will do it by itself. Not yet, unless we get so advanced like in all those movies. And you don’t know what advanced tech is out there, that is not released to the public. You acting all high and mighty, like you know everything about this is fucking disgusting.
If you mean that LLMs are more powerful brute force attackers, sure, but “Anthropic models hacked” and the OP you wrote yourself is just fearmongering. At best LLMs can tech a person who did not previously know how, how to commit certain crimes, and possibly create bash scripts to automate this, but that is not a model that autonomously hacks.
The hype is purely to drive the coming IPO, they have to go hard and fast before the world realises what rouse is being pulled.
If you just want to critically think for one moment: (leaving asside models have no autonomous agency) If the Mythos-class models really are that impressive, and it did what they claimed, how was it not disabled before Fable went public? Mythos had been ‘out’ for over a month to those select few orgs.
There’s a discussion in the House of Lords last week which is a bunch of grandma’s and grandpa’s yapping on about the dangers of AI. You remind me of them. Remember how little people understand about the toxic algorithms of social media? They understand about 100x less of this, and are all making kneejerk decisions because they don’t understand anything and are scared if something goes wrong, the buck will stop with them.
And Anthropic and Open AI both benefit from not educating people too much, because that would pop the Myth and the Fable they are trying to sell.
Yes, I am talking about the tools used to break in systems. And wait a minute, do you actually think that ai is not a threat? You are acting like you know about every unreleased ai related thing out there. If these systems were to exist, no way it would be released to the public. You don’t know everything about it and I don’t either.
I have an above average understanding of the underlying technology, and there is no more threat today from “AI” than there has been from other sophisticated tools circulating on the dark web for decades.
To use an analogy: I understand how a combustion engine works. If you suddenly came out and claimed that combustion engines are able to mow grass and we would never have to mow lawns ourselves gain, I would understand enough to know you’re out of your depth.
You can believe pigs can fly, but to people who understand pigs, they will just see you’re talking about beliefs, because the knowledge and the science says, that simply because of their lack of wings and fairly dense bodies, the rotation of their tiny curly tails is simply insufficient to provide the lift to make them airborne.
What you can do is put lipstick on said pig, and that is what I’ll let you do now, just expect to run into resistance when you talk to people who know instead of believe.
So you think you know what all the CIA or FBI has, lol. Ok, got it.
No I didn’t claim that. I’m saying that whatever they have that you believe and are speculating is dangerous, isn’t LLM based, because that tech is incapable of from its technological foundation to achieve what you’re trying to fear monger.
You don’t know what is to come with it. Quit acting like you know all the capabilities of ai everywhere.