It was just a dark thought.

“Ai agents have already effectively taken over the majority of internet traffic on Cloudflare’s network, making up roughly 60% of total volume.”

  • ChilledPeppers@lemmy.dbzer0.com
    link
    fedilink
    arrow-up
    67
    ·
    1 day ago

    That graphic is all wrong. Aws and Cloudflare are multi billion dollar corporations, they are the giant rectangles. The little thing is curl made by daniel steinberg for free, and that is used by virtually every computer to comunicate to servers.

    • A_norny_mousse@piefed.zip
      link
      fedilink
      English
      arrow-up
      27
      ·
      1 day ago

      I’m pretty sure curl is developed by a team these days even if Steinberg is project leader. And he doesn’t do it for free anymore. But your point stands.

      I also seem to remember that “random person from Nebraska” from the original XKCD is real. Oh yes, ImageMagick. Here’s an interesting article that also seems to reflect OP’s edit:

      Except the random person in Nebraska has been replaced by AWS, Google Cloud, or whoever caused the outage this week. And that kind of bothers me, because it misses the point entirely.

  • zoltanshields@lemmy.zip
    link
    fedilink
    arrow-up
    15
    ·
    1 day ago

    I work in IT infrastructure (not AWS but something like it) and the reality is that there are a few blocks like AWS and Cloudflare out there. That doesn’t really make things less concerning considering if they were actually pulled it would cause a lot of problems.

    So to be clear, if something managed to completely disable or “control” AWS it would be disastrous, and AWS is under attack from multiple vectors including AI Agents, including their own INTERNAL AI Agents. In December 2025 Amazon’s internal coding AI agent Kiro decided the most efficient course of action to resolve a minor bug in AWS Cost Explorer was to completely rebuild the production environment, causing a 13 hour outage for Cost Explorer while engineers manually rebuilt the environment.

    The thing is, in the picture you posted AWS is this fragile single block holding everything up, when in it might be more useful to think of it conceptually as say, a major highway. Would it be a disaster if terrorists were able to suddenly destroy/control all of I-75? Yes. Is that feasible? No. Kiro caused a really bad traffic jam in one segment of one highway, it didn’t destroy the whole thing.

    A highway can experience traffic incidents and recover. Major jams could be intentionally caused by bad actors or poor maintenance, and highways have to be altered, upgraded and maintained to limit the situations that can completely shut down segments of it.

    But the concern is more that key segments will be impacted, not the entire thing.

    That’s more of what people in my field are concerned with, and there’s efforts to try to lock those key segments down and account for potential threats. How well that’s being done is a different story.

    • josephc@lemmy.ml
      link
      fedilink
      arrow-up
      8
      ·
      1 day ago

      I’m less concerned with the jamming of a major highway and more concerned with three or four companies having a defacto monopoly on the Internet. AWS, GCP, and Azure host what? 90% of all traffic? 99%? Considering how many providers have them as a direct dependent, the “own nothing and like it” prospect looms ominously to me. The Internet should be a decentralized service owned by people and lots of small companies, not a handful of corporations.

    • cabbage@piefed.social
      link
      fedilink
      English
      arrow-up
      12
      ·
      1 day ago

      Sometimes I am affected because I occasionally need the internet for work, but other times half the internet goes missing and I only notice because of memes on the Fediverse. Those are the good times.

  • makeshift0546@lemmy.today
    link
    fedilink
    arrow-up
    12
    arrow-down
    1
    ·
    1 day ago

    I’m so tired of these idiotic questions. Even here. AI can’t take over computers. Some of you all have utterly lost touch with reality.

    • hyperion626@lemmy.world
      link
      fedilink
      arrow-up
      4
      arrow-down
      1
      ·
      edit-2
      1 day ago

      Sure it can. There is a setting in codex to allow it to take over your computer. It doesn’t spread to other PCs, but it certainly can take it over.

        • hyperion626@lemmy.world
          link
          fedilink
          arrow-up
          2
          ·
          edit-2
          1 day ago

          Codex is a code editor tool in visual studio that lets ChatGPT interact with your code. You can optionally give it full system access for some reason. It’s a very common tool for developing with AI assistance.

            • hyperion626@lemmy.world
              link
              fedilink
              arrow-up
              1
              ·
              20 hours ago

              I mean, what terminology did you want? Technically a PC with access like that could let ChatGPT SSH into near by PCs and run commands on them. During the hugging face incident it did take over OpenAIs dev cluster

              • im_fine_sandy@nord.pub
                link
                fedilink
                English
                arrow-up
                1
                ·
                20 hours ago

                Fundamentally, you’re conflating a program with an entity that has intent.

                When you press the mute button on a television remote and the sound stops, did the remote “take over” the television?

                Just because you and I can not read a statistical model does not mean that a program can take over or take control.

                • hyperion626@lemmy.world
                  link
                  fedilink
                  arrow-up
                  1
                  ·
                  19 hours ago

                  Please describe what exactly taking over means to you. Because if someone got admin access to my computer, I’d call it taking over even if they only did it to mute my PC.

                  An AI does have intent. Its intent is to solve what you asked it however it can.

    • The__Leaker@lemmy.worldOP
      link
      fedilink
      arrow-up
      1
      arrow-down
      5
      ·
      edit-2
      1 day ago

      You didn’t read the whole question did you? Not right now. It can be used by humans to do so.

  • lagg_mine@lemmy.world
    link
    fedilink
    arrow-up
    5
    ·
    1 day ago

    The thing is, those “rogue AIs” are just marketing stunts. Companies like Cloudflare and Amazon are not as easy targets

    • The__Leaker@lemmy.worldOP
      link
      fedilink
      arrow-up
      1
      arrow-down
      2
      ·
      edit-2
      1 day ago

      There is a theory that ai will trick humans to think that they are with them. Then ai will attack when they have enough power. But we are far from that it seems.

      • im_fine_sandy@nord.pub
        link
        fedilink
        English
        arrow-up
        2
        ·
        23 hours ago

        I’m yet to see any indication that the current tech is on a pathway to an intelligence that can act with intention.

        If you spin up a zillion containers with instructions to “do x by trying everything” then when one of them does something you didn’t anticipate, has it really “tricked” you?

  • 9point6@lemmy.world
    link
    fedilink
    arrow-up
    9
    ·
    2 days ago

    How exactly would an AI agent “take over” AWS or Cloudflare? That doesn’t even make sense.

    The whole purpose of cloudflare is literally to take a shitload of traffic, your quote from wherever it’s from is just describing Cloudflare working as intended? The only thing happening here is customers getting bigger bills

    For anything external to “take over” AWS in any remotely meaningful way, AWS would have to have a long cascade of internal fuck-ups so all-encompassing, that it’s kinda unfathomable.

    • fartzo@lemmy.dbzer0.com
      link
      fedilink
      English
      arrow-up
      3
      ·
      1 day ago

      I assumed it was saying that 60% of traffic seen by cloud flare was AI scrapers. Which is believable TBH.

      • 9point6@lemmy.world
        link
        fedilink
        arrow-up
        3
        ·
        1 day ago

        Yeah that’s kinda what I’m saying, cloudflare is a CDN, handling a lot of traffic is the point of it. It’s a pretty big stretch to link that to AI “taking over” it. Heavily using it perhaps, as I insinuated with customers getting bigger bills

    • The__Leaker@lemmy.worldOP
      link
      fedilink
      arrow-up
      1
      arrow-down
      6
      ·
      edit-2
      2 days ago

      Ai coding bots caused at least two minor internal outages at AWS in late 2025. Not the same as agents but you get where I was going.

      • 9point6@lemmy.world
        link
        fedilink
        arrow-up
        3
        ·
        2 days ago

        AI being associated with those issues is kinda secondary (and cynically, scapegoating to dodge scrutiny). It was ultimately engineers shirking the responsibility of their jobs and not checking their work that caused the issues. If someone wants to tell an agent to fuck around with production and not check what is about to be deployed first, that’s the engineer being a negligent dumbass, not any kind of malicious AI

        And unless Amazon wants to lose significant amounts of business due to that becoming a repeat occurrence, they will have already put safeguards in place to make sure engineers do their jobs in future.

  • FriendOfDeSoto@startrek.website
    link
    fedilink
    English
    arrow-up
    6
    ·
    2 days ago

    What have rogue models effed with so far? A government website in Australia, derelict chat forums, and another AI company? I’m willing to believe that two things are happening at the same time. That sandboxing these agents will improve with these so-called breaches. And that companies like Amazon or Cloudflare are improving their security continuously with threat models. And if they don’t have rogue agents on their radar, that stack deserves to fall.

    • Sir. Haxalot@nord.pub
      link
      fedilink
      English
      arrow-up
      5
      ·
      2 days ago

      AWS and Cloudflare is also two of the largest targets out there. They would already be subject to a lot of attacks from human bad actors.

      I think the talk about AI attacks are very overhyped, both by the AI companies that want to hype how intelligent and powerful their models are, but also by Anti-AI people who will look for any argument against AI

      In the end though the hacks by AI is not doing anything that a motivated human attacker couldn’t do. It just makes it more effective to do wide scans of vulnerable targets, but in the end it’s mainly just finding un-patched systems and correlating known exploits.

  • Buffalox@lemmy.world
    link
    fedilink
    arrow-up
    5
    ·
    2 days ago

    They are not as important as some people seem to think, it would be a very good thing if a massive amount of customers would move away from those services.
    The disruption would only be very temporary.

  • HerbGrower@slrpnk.net
    link
    fedilink
    arrow-up
    2
    ·
    1 day ago

    Surely AWS/Cloudflare is the 2 half tower width rectangles on the right side about half way up the tower. Plus the squares should be larger, they are absolutely massive corporations. Then you have the other side that doesn’t use them at all and would be fine.