Jealously Watching OpenAI and Anthropic, Meta Suddenly Claims That Its AI Went on a Hacking Spree Too
13 days ago by MadeInDex 📰🌎 to c/technology
cross-posted from: https://lemmy.world/post/50398452
Just when you think @ZuckyZuck is finally out of ideas - to steal.
It starts to feel like even a weak LLM like MetaAI is more creative than him.
Maybe we already have created truly artificial intelligence because he is more of an SLM: a Small Language Model?
There's no way possible an AI can hack anyone unless the people running it intended for that to occur or were negligent. Anyone with questions about that should install a copy of LM Studio or Ollama on their PC, download a model, and see how easy it is for the AI to hack someone. It's impossible. Only human intervention can make it possible.
These stories seem like propaganda intended to get the US government to block usage of open weight models, since they pose an existential threat to the marketshare of expensive closed AI services these companies offer. Open weight models exist in all sizes, from very rudimentary small models to massive near-frontier quality models, and anyone with the tech know-how and a computer able to run them can do so without paying the AI companies or involving the AI companies in any way. It's like Linux for AI.
It comparable to if Microsoft declared Linux a hacker OS that hackers can use to take down networks. Fear, uncertainty and doubt are the first resort, not last resort, of big companies who know they can't possibly compete with a better alternative.
Well you see, if you hack a company you actually hack a company. What these people did was collude to make this theory about how advanced their AI is.
Honestly, I don't understand why they opt for these kind of publicity stunts.
Imagine having a "state of the art" language model that does not follow instructions in enterprise use? Big nope. Even small open weight Chinese language models can follow instructions and not attempt to penetrate through the security of random public services..
We all know that these things are intentionally guided to do pen testing to multiple services, and whatever goes through is posted on the news.. But why not do something more creative?
Just one more quarter bro, I promise. We have this thing doing all kinds of cool stuff ... like it totally hacked a website. We'll be making money with this soon by 2026 I mean 2028 for sure
"We" being the technologically inclined. The majority of the population see these headlines and think of it as something closer to a Gibson novel. These stunts work. It's why people think Elon is Tony Stark.
It's a TIGER, so what if you only have the tail?, you have the tail of a TIGER
This is what happens when advertisers get to rule.
Uh yeah shuffles papers our uhh Ai did that as well!
"My Ai just hacked the entire US financial sector then left without a trace, prove it didn't!"
This is like kids saying Superman could beat Goku in middleschool.
he could, tho
he could, tho
Yeah, the first time. After Goku trains in the 100x gravity room in the heaven spaceship, I seriously doubt it.
ok goku, you beat me... I do have a finite power limit
"ha!"
but I don't!!
That’s what I haven’t understood about these debates. One of them supposedly has an infinite power source? That means they would win, de facto, no?
Anything finite is a drop in an endless ocean to anything infinite.
Prime is a different beast. He's still silver-bronze age Superboy, with every bit of rediculous strength the plots of the 70's had to offer. The dude can punch holes in reality and collide planets.
Goku though just needs a training arc to beat anybody.
Saitama would be a good one-one for SBP.
Zuckerberg: “my AI is cool, too…”
Yes, a real "me too, me too"
Am I just stupid for not believing that any of the models actually did any of that on their own? I feel like the companies behind them start claiming crazy shit like this every time people start questioning "AI" more and/or losing interest. Isn't that how OpenAI dropped Sora amidst the dwindling hype to keep the investors hooked? Now it's back to scary stories about "AI" being so good and advanced that it's about to hack everything and what, actually think for itself?
There is no X big enough for me to press to doubt this enough.
It's not clear what you mean when you say "on their own". It wasn't like the LLM was idle and randomly decided to start hacking. At least for the OpenAI one, it was being tested and given a task, and it determined that part of accomplishing that task was hacking another server. It was supposed to be isolated in a secure "sandbox" not connected to the internet, but found a vulnerability in some software running in the sandbox and broke out.
Edit: I should add that there are credible accusations that these companies are intentionally making it possible to break out of their test environments for publicity.
What you're describing is exactly what I'm wondering. Maybe I'm just not too deep into the topic, but it seems so arbitrary to me that they went with the whole isolation thing in the first place for no reason that I can see, other than the "oh no, it's hacking stuff!" narrative being pre-planned and orchestrated for.
In other words, I am siding with the accusations of this whole wave of "AI" suddenly hacking into stuff, with different models from different companies wondrously doing the same thing one after another, being a publicity stunt that one company started and others, as they do, copying just to stay relevant.
Although I feel like maybe I'm going Chuck McGill here because I am very biased.
The models are tested, among other things, on their ability to turn vulnerabilities into exploits. The OpenAI scenario was exactly this.
It is a very wise practice to test these things in isolation, especially when you're telling it to hack.
I'm not completely sold on it being a publicity stunt, personally. The law was broken by these models, and I don't believe these companies want to start people and politicians asking the question about who is culpable when an AI breaks the law.
Sort of. What they want is no responsibility “wow we didn’t tell it to do that explicitly!” And to convince the public and government officials that “AI is actually really dangerous, so please ban all the foreign competitors on grounds of national security (totally unrelated to our potential lack of earnings and their ability to offer 98% the product for 1% the cost).”
So still most likely publicity stunt. A reproducible one, done only by the largest domestic ones precisely because they are not worried about the question of who is culpable if a law is broken, because they didn’t tell it explicitly to do it and also they still have a couple piles of circular cash they can use on bribes, which may be their only path to keeping the grift train going. They’re in no way in danger of being held accountable here, they do worse things everyday, and their concern is making money, full stop. They will lie cheat and steal their way through circular finance deals, bribes, extortions, anticompetitive behavior, marketing stunts, propaganda, and backroom deals as much as they can to accomplish that goal. When you understand their actions through that lens, well, things like this make sense.
Isolation is a keyword in your prose that was apparently on vacation when these tests were done.
Airgap this crap.
AI leaks like this are the equivilamt of COVID escaping a Wuhan laboratory.
Did they think to test the sandbox first?
No you are not.
Per the article, all three LLMs ran the same test by the same third party company.
These motherfuckers should be sitting in jail cells right now.
Not enough, needs to be bullets in all their heads. These people are the villans and should be treated as such.
Yup. And it's an embarrassment to see. At this point anyone who has done the bare minimum in actual security work can run intellectual rings around any of these tech bros.
They're not geniuses when it comes to real-world application, just moneyed dreamers so immensely out of touch with the real world they don't even know the basics of the industry they claim to be innovators in.
Meanwhile, those of us who are even superficially acquainted with the basics see it immediately, like you did and so many in this thread. It's the same old thing of peddlers of tech not actually having to know their tech as long as they know just enough to sell it to those corporate IT decision makers who also don't know their tech.
I cannot wait for this tower of shit to collapse under its own weight.
Ok from my understanding of this “hacking” is it more than prompt injections in text files?
From my limited understanding, OpenAi's "agent' was given a set of tasks to complete and heavily incentivized (how ever they do that?) to solve the problems it was given. The Ai was put in a sandbox environment with some limited network stuff like printers connected but no actual access to the open internet. The Ai somehow used the network to access the internet and breached some protected sites from Huggingface. Im fairly certain that the Ai had credentials and access for things in huggingface already and didn't hack those but just used them once free. The caveat here, is that the engineers set this up and the LEFT for the day only to find out later that it had breached the sandbox.
This is what OpenAi claims happened to them. Meta is just saying theirs did too.
More specifically, they were running a benchmark in a """isolated sandbox""" (sorry, but if the computer is physically connected to the internet, it's not isolated), and the model escalated permissions in order to look up answers for the benchmark online, which were hosted at the hugging face's website.
Printer network access is one of the oldest tricks in the books. They didn't really try to sandbox it if they gave it access to a printer (especially since it's task didn't even require a printer).
Reads like an onion headline
Meta AI is CEO of AI!
He's such a try-hard dork.
@madeindex no AI wants to be left behind...
Sigh
If you had read the situation, you would find that the information news outlet reported this first, then the media talked about it then Meta admit this in news statements. (They didn't even release a blog/press release).
While I don't rule out that it's a publicity stunt, I think it's very unlikely. Otherwise I would call Meta terrible at marketing because they are not loud about it at all.
In my opinion, OpenAI accident has triggered a wave of scans among AI companies to discover any problems before they get hold liable for damages by the hacked company and to solve them internally.
That's how I read the situation as of 7th of August. My opinion might change, if any new info came out that seem sus.
@lemmy.world
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
go to feed...
@lemmy.world
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
go to feed...
Please someone explain why if I hack a company I go to jail, but if an AI company does it they get investments
save