"Anthropic’s AI Model Hacked Nearly All NSA Classified Systems, Chief Claims."
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.
https://www.tipranks.com/news/anthropics-ai-model-hacked-nearly-all-nsa-classified-systems-chief-claimsOpen linkView original on lemmy.world
X Doubt
If it's so good, where are the Epstein files?
You think the government would actually release all of them anyway? Think about what you are saying, the NSA has it. This is the NSA we are talking about.
What?
I'm saying if the AI model is so good at hacking government systems, why haven't they obtained the Epstein files?
Of course the government wouldn't release them. Bondi even said that if they did the system would collapse from how corrupt they all are. Of course they'd never release it of their own volition, that's why we've got a textbook limited hangout, they're hoping people will be satisfied with the less disparaging info and forget about what they haven't released.
Meanwhile, Mythos is getting close to being replicated on open source
https://medium.com/open-intelligence/openmythos-the-open-source-reconstruction-of-claude-mythos-that-reframes-what-ai-scaling-actually-32297d4be231
So my point is, if it's so good at breaking into government databases, where are the leaks?
Or is this just a publicity stunt for Anthropic?
I mean... There are a lot of reasons to be skeptical of mythos, but taking this situation in a vacuum, the hack against the NSA was done by a team at the NSA, for pen testing purposes. Why would that lead to them releasing anything?
What I am saying is, it hasn't been released. Has it? Let's see if Mythos actually does. If the some government entity in the USA are the only ones that have it, they won't be hacking the Epstein files. Unless it is some rogue employee and it is actually capable.
Fable/Mythos was already shown to not be that much better at vulnerability finding than previous models. If anything this is basically saying that security at the NSA really isn't up to par.
It's too dangerous !!! Btw buy our stock, our IPO is coming soon
So it can be made more dangerous...
Says who, anthropic?
It was tested to see if it can break into systems and it "did".
"Chief claims" - without evidence, this means absolutely nothing. Especially as this was a higher up, who likely heard about something happening through multiple layers of reports.
I still believe that there will be more dangerous ai models, and the guardrails will be removed.
The key word there is "believe".
Until there's evidence, that's all you have: a belief.
Ai will become more dangerous. I am telling you, it is not just about believing. It is and it will continue. It will mess up and make people crash cars for example.
The only danger this round of AI will bear is fiscal. When, not if, the bubble bursts, one country in particular is going to have a really bad year economically speaking.
The bullshit market hype that Anthropic is putting out about how "dangerous" its "latest model" is ... well, it's exactly that. Market hype.
It isn't hype if some ai is already causing problems. Examples: Deepfakes, misinformation and cyber security threats. A model was already yanked from the customers.
Completely baseless and stupid claim from an admin that lies more often than not. This would be crazy if true though.
If you actually think that ai can't hack systems already, I got some magic beans to sell ya.
Hey buddy... How does AI work? Just, briefly explain how it works so we can verify that you know what you're talking about.
There is no "AI". It's just a computer program. I imagine it's largely fuzzing optimized with statistics, methodology, etc. This approach seems legit, although I can only speculate on the efficiency. I guess these grifters have never cared about efficiency, esp when hacking the NSA.
That is a broad question... With that question, It looks like you are asking because you don't know. What kind of ai? If you are talking about ai systems, they learn from huge amounts of data, and it identifies patterns to make predictions or decisions.
Oh, no, I know how these things work generally, but do you? I'll let you decide if you wanna pick LLMs or Generative AI or Machine Learning. How does any AI "learn"? How does it do anything? How would any of these AI systems hack anything. Just... Walk us all through it. Because if you don't know how these things work or would work, then I'm not gonna trust your assertions.
It learns through pattern recognition. Did I say it can do it on its own?! If it looked that way, I was meaning in the future. Ai tools can be used by a user now. A user, not hacking by itself, or on its own.
If you think its hacking nearly all NSA systems you're cooked.
Like someone said in the comments, don't underestimate ai. That is how it will kill us all.
Hey look it's a 5 day old AI evangelist account.
You clearly don't understand the technology, which disqualifies you from fear stoking the masses who also dont understand it.
You clearly don't know who I am and just assume that I don't know anything about the tech. Based on a newer account. How tragic. You sound like one of those Reddit bots. Yuck!
You believe AI can hack systems, that's what i'm basing this on, and your other comments which are all promoting LLMs to be far more than they actually are.
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.
A dipshit LLM isn't gonna kill anyone directly idk what you're on about lol
The terminator is not real
I was talking about overall in the long run. Jesus...
You really bought into their advertising huh?
Ai will take over, but not this.
A computer program developed to find bugs is good at finding bugs?!?!? Shocking.
Zero "AI" involved.
Mythos is literally an AI model...
I think he does not define LLMs as AI which I would agree with.
Pump and dump?
Not much time left, yes.
https://apnews.com/article/tech-stocks-ai-investments-8a0ff4c95d5cae6f65c6e2ba03047058
It's possible.
All I see is market manipulation. I'd be happy to be proven wrong when it's released worldwide