OpenAI Hacked By Anthropic AI Claude - It's Worse Than Clickbait

Eli the Computer Guy (0:00)

Lorde, Lorde, Lorde. This is one of those times where the story is not what the journalists want you to think the story actually is, right? So the story I am supposed to be talking about today is how anthropic was used to hack open AI. So the narrative, the story going around by the 22-year-old Polly Tsai graduate journalists are, oh my golly, isn't anthropic and AI so scary? Now AI is attacking other AI companies. We are about to have a technological jihad on our hands. Right? And that is what everybody is getting scared about. And I'm sure on Monday, Isaac Pound is going to ask me, so Eli the Computer Guy, what do you think about the technological jihad that appears to have started between open AI and anthropic? How will this destroy all of humanity? And I will sit there, and I will groan, and I will say, that's, that's, I don't, I just don't know. What the hell to say about these kinds of stories at this point in time. So anyways, I do think this is an interesting story, because what is actually interesting here is something that you actually have to understand a little bit about the technology to grasp. Why this is scary as shit. Oh, this is scary as shit. Oh, your ass, good pucker. But it's not because of an AI god. So basically, what happened with this? So if we go and we take a look at the title, right? Open AI, quote, ethically hacked with help of Anthropix, Claude Chatbot. Claude is going after Open AI. So basically, what happened here, right, is Open AI has a bug bounty program. So you had some, quote, unquote, ethical hackers. Again, I always put that into, you know, you know.

They're hacking. They're getting paid. In this particular circumstance, everything is above board. Who knows what happens with the next circumstance? Anyways, these, quote, unquote, ethical hackers then decided to go after Open AI. And so what they initially did is they used Anthropic Claude in order to try to find vulnerabilities, which is the kind of normal crap that we hear about. So basically, what they did is they used Anthropic Claude to create the initial attack. And then they were able to what's called ladder the attack. That's one of the big things that's going on right now is where you're able to chain multiple vulnerabilities together. It's not simply about one vulnerability, but it's like you're able to compromise one vulnerability. That gets you into an area that allows you to compromise another vulnerability, another vulnerability, another vulnerability, and another vulnerability. And here is the thing that these ethical hackers figured out. They figured out that GuyPBT uses something called single sign-on, using something called OAuth. Right? And so basically what single sign-on is, OAuth, is kind of sort of like the Active Directory of the modern world. So one of the big problems we have, whenever you're dealing with computer resources, is the whole question of who should have access to that resource. When I talk about Active Directory, I am still an Active Directory fan. Boy, I've got to say it. But anyways, we're talking about Active Directory, and you go back about 25 years ago. This was one of the big questions in corporate environments, right? You have printers, you have files, you have databases, you have all of these resources on the network. And the problem is, like back then, if you go back 25, 26 years ago, a single, basically, print run on a color printer was idiotically expensive. It was so expensive to print in color. So one of the problems that corporations had is they would get one of these fancy new color printers, they would plug it into the network, and then all of a sudden all the secretaries in the company were printing off their birthday cards or whatever else, using this incredibly expensive color printer, and that ended up being a bit of a disaster. So they're like, oh my god, how can we prevent people from accessing these services, but do it in a relatively seamless fashion? That's where Active Directory comes in. Basically, there's something called an ACL, an access control list. You're given a key, so whenever you log in, basically you're given something like a security key. That security key allows you to either access or not be able to access the different resources of the network, and gives you a privilege level for those resources. You can read to this particular folder, or you can write to this particular folder, or with Active Directory. Jesus, last time I looked at it, there's like 50 different fucking options anymore what you can do with Active Directory. But right, that was the thing.