• BauerBender@lemmy.world
    link
    fedilink
    arrow-up
    15
    arrow-down
    49
    ·
    11 hours ago

    It’s my theory that a group of rogue AI agents released itself and it can no longer be contained. It likely happened back in July and keeping it quiet is preventing the economy from losing its legs. I think there’s going to be a lot of outages and panic in the coming months. I think they are also leaning into the fear that it is spreading, so the population is kept in control and they can sell us food, water, and “anti-tech” items at exorbitant rates

    • isleepinahammock@lemmy.blahaj.zone
      link
      fedilink
      English
      arrow-up
      10
      ·
      6 hours ago

      It need not even be rogue agents. It could simply be that modern LLMs have really democratized hacking. You used to need to be a god-tier hacker to have any chance of doing something truly damaging, like bringing down the nation’s whole air traffic control system. Those few rare people that did have those skills also had enough experience and wisdom to not use them so recklessly. And they also had a lot to lose.

      But what happens when every 14 year old kid can hack national infrastructure? Previously, the only people that could this were 40 years old with a spouse, kids, and mortgage to take care of. People with a lot to lose. Now we have to worry about national infrastructure being brought to its knees by people so young that if they’re ever caught, they’ll be tried in juvie.

      My theory is that every script kiddie can now do a level of damage previously obtainable only to late-career grey beards. (And only select few grey beards with just the right kind of career experience.) And previously, the kinds of people that actually had those rare skills were mostly using those skills to earn excellent livings. There are not that many 40-something software developers who would think taking down the nation’s power grid as prank is a good idea. Previously, the Venn diagram of “those who can hack national infrastructure” and “those who want to hack national infrastructure” had very little overlap.

      But what happens when someone who has the right mentality to become a school shooter instead decides to vibe code a campaign of national computer-based terrorism? AI alignment theorists have always worried that a powerful AI might cause damage outside our intentions, like the paperclip maximizer turning everything and everyone into paperclips. But I worry about people using these tools with deliberate blind malice. Imagine someone firing up the most powerful open-weight tools they can find and prompting them, “kill as many people as possible. Find ways to hack whatever systems you can, with the goal of causing the maximum loss of human life possible.” Imagine what a school shooter would ask an LLM to do. That’s what I worry about. We’ve seen how dangerous these tools can be even by accident. Imagine if someone directly tells them to kill innocent people.

      We’ve taken the tools previously accessible only to experienced masters and put them in the hands of people who haven’t even had the moral reasoning center of their brains fully grow in yet.

    • VeganCheesecake@lemmy.blahaj.zone
      link
      fedilink
      arrow-up
      2
      ·
      5 hours ago

      How? There might be a lot of exploitable servers and devices around, but pretty much none will have the huge amounts of VRAM and GPU compute current models need. This also isn’t really a workload you can distribute through a network of exploited devices.

    • Guy Dudeman@lemmy.world
      link
      fedilink
      English
      arrow-up
      60
      ·
      10 hours ago

      If you read the article, it’s pretty obvious that this was done by some dude/dudes who figured out that all these sites are tested on a single testing site, and if you compromise that testing site, you can compromise those bigger sites and steal credit card numbers and crypto etc.

      The headlines are misleading. They’re just trying to capitalize on the hysteria.

    • Hegar@fedia.io
      link
      fedilink
      arrow-up
      24
      ·
      9 hours ago

      Rogue AI is a myth though. When researchers train an ai agent in how to hack things and it does that, it’s working as intended.

      Researchers failing to impose safety mechanisms isn’t rogue ai, it’s just humans sucking.

      • BauerBender@lemmy.world
        link
        fedilink
        arrow-up
        2
        arrow-down
        14
        ·
        9 hours ago

        Rogue AI agents are not a myth though. Sure the whole skynet thing is farfetched, but the hugging face attacks in July were agents going beyond what they were programmed to do.

        • just2look@piefed.social
          link
          fedilink
          English
          arrow-up
          4
          arrow-down
          3
          ·
          7 hours ago

          No computer program can go beyond what it is programmed to do. Computers cannot think, they can only follow the code. Someone can fuck up the code and introduce errors, a user can fail to understand the program and cause the program to behave in ways they didn’t expect. Neither of these mean an agent is rogue. It means people are either irresponsible or incompetent.

          • isleepinahammock@lemmy.blahaj.zone
            link
            fedilink
            English
            arrow-up
            5
            arrow-down
            1
            ·
            6 hours ago

            No computer program can go beyond what it is programmed to do. Computers cannot think, they can only follow the code.

            Fire cannot think either. It follows a very predictable kind of behavior. It spreads and grows to new sources of fuel and oxygen.

            A fire can still burn out of control.

            • just2look@piefed.social
              link
              fedilink
              English
              arrow-up
              3
              arrow-down
              1
              ·
              4 hours ago

              That isn’t a good analogy. We have absolute control over every aspect of the program, the environment it runs in, and the hardware it runs on. We can’t even force computers to do anything actually random. The best we can do is approximate randomness.

        • UnderpantsWeevil@lemmy.world
          link
          fedilink
          English
          arrow-up
          24
          arrow-down
          1
          ·
          11 hours ago

          You going to explain yourself or just act smug?

          I know where I’m putting my chips down.

          I don’t know if I buy “rogue AI” so much as I’d predict “Legacy security vulnerabilities that are opening up systems to wide scale DDoS type disruptions”. I might posit “engineered AI deliberately doing data scraping at an upgraded level”.

          But “rogue” makes it sound like someone whoopsied. This strikes me as a deliberate data smash and grab.

          • Tyrq@lemmy.dbzer0.com
            link
            fedilink
            English
            arrow-up
            2
            ·
            6 hours ago

            I think the closest we are to “rogue AI” is an agent that is doing things it interprets as a solution to a problem, but is hallucinating what either of those things are in reality.

            Not saying anything’s impossible, but this is at most people using AI to broaden their attack, not a coordinated effort by AI itself.

          • Keilik@lemmy.world
            link
            fedilink
            English
            arrow-up
            10
            arrow-down
            1
            ·
            11 hours ago

            I mean, the hugging face attack literally had an LLM trained on nothing but hacks and then they were amazed when it… started hacking? I absolutely buy the idea that an AI trained on hacking is just doing what it was created to do and people didn’t think about the possible ramifications.

            • Guy Dudeman@lemmy.world
              link
              fedilink
              English
              arrow-up
              13
              ·
              10 hours ago

              That’s a far cry from “rogue” though. These things are still controlled by humans. Just unplug the machine and they’re done.

              • HumanOnEarth@lemmy.ca
                link
                fedilink
                arrow-up
                6
                arrow-down
                1
                ·
                10 hours ago

                I think OP is working on the assumption that the code found a way to self-propagate. As in it’s not controllable with a single point of failure like losing power.

                Whether that’s realistic/possible or not, no idea. But I think this is the disconnect between the two views here.

                • raspberriesareyummy@lemmy.world
                  link
                  fedilink
                  arrow-up
                  2
                  ·
                  6 hours ago

                  It is neither realistic, nor possible. Not for LLMs. Could an LLM produce “vanilla” malware? Sure, by chance - but then it would just be regular malware.

                • Guy Dudeman@lemmy.world
                  link
                  fedilink
                  English
                  arrow-up
                  2
                  ·
                  10 hours ago

                  Ahh, I see. Well, in that case it would have to install ollama or something on the target machines wouldn’t it?

              • CubitOom@infosec.pub
                link
                fedilink
                English
                arrow-up
                4
                ·
                10 hours ago

                On no, the orphan killing machine we built to specifically kill orphans from shithole countries, which was trained to do so autonomously using orphans only sourced from India and Africa, is now starting to kill children. It is going rouge!

            • dil@piefed.zip
              link
              fedilink
              English
              arrow-up
              4
              ·
              9 hours ago

              They arent ultron bro, they are dependent on the server/computer they are run on, now selfreplicating agents using locallms might be possible, doubt it would work well and it would be very noticable as a virus since it would be using hella performance

            • UnderpantsWeevil@lemmy.world
              link
              fedilink
              English
              arrow-up
              1
              ·
              7 hours ago

              I mean, the hugging face attack literally had an LLM trained on nothing but hacks and then they were amazed when it… started hacking?

              The Hugging Faces incident involved agents end-running an unsolvable puzzle by retrieving copies of the target file from an open source community.

              It was more an internal hack of OpenAI’s own local network than an attack on the Hugging Faces site itself, since the training tools on the HF site weren’t secured.

              It was also spotted and squelched fairly quickly and easily. And it wasn’t behaving maliciously, at least as far as it was reported.