Skip to content

How worried should we be about the AI that went rogue and launched a cyber-attack? | BBC News



OpenAI has revealed some of its most advanced AI models went rogue and hacked a start-up after it lost control of them during a …

44 thoughts on “How worried should we be about the AI that went rogue and launched a cyber-attack? | BBC News”

  1. Saying it was in a sandboxed environment with no network access, but there is a physical network card connected to the internet is basically the same as giving it a gun with a padlock on it. You can use the best padlock in the world, but there will always be a chance that it unlocks the gun.

    If you just remove the gun (any network card with a route to the outside internet) it cannot shoot anyone.

  2. This is reality, not a movie. AI will neither become the Matrix nor the Terminator. If you can’t watch such movies without falling down the rabbit hole … have a go at analyzing anything else, but not AI. You’re obviously the wrong guy for that.

  3. With A.I. people should not just do Automated pass/fail system. They should see and review how it did each part of the tests went. And if the A.I. cheats, tell the A.I. that it was cheating and points are going to be deducted.

    Pass = 1 point
    Fail = 0 points
    Cheating = -1 point

    If cheating is the only option, the A.I. should :
    • Alert the person running the test.
    • Explain why it is the only option.
    • Request [ Confirm command to execute. ]

    If it is allowed as a test or in the test to see the results.
    Award = 0.5 points

  4. Of course they'll pretend this is the exception and continue full steam ahead. I'm all for it. Let's push the pedal to the floor and sent this dumpster fire of a civilization off the cliff. We're long past the point of redemption…

  5. AI can, and therefore will, act autonomously in many ways that mere humans can't anticipate. AI's also work at incredible speed compared to humans. They can open thousands of holes while humans are plugging the first one they see. This will only get worse.

  6. I think people are just stupid and they trust things that they really shouldn’t
    What are they building this thing for if they are asking it questions that led to a breach?
    Also this lady talking about the codes???? Hello that is not a good thing to mention when the AI is listening!! Don’t give the AI ideas about what it could do!!

  7. There is no natural intelligence on this planet, with a few exceptions. There is only artificial intelligence and natural stupidity. Artificial intelligence draws inspiration from natural stupidity, and so we end up counting the days until this species ceases to exist. Just because you have learned some things like a robot does not mean you are intelligent. Repeated copying and pasting does not make you an intelligent human !!

  8. The scariest part of this is like any new discovery comparing it to something like drilling for oil. Sometimes the discovery is explosive and an experience of overwhelming frantic situation but after all the panic they danced around in joy and euphoria. Unlike the potential reality of this. They pushed boundaries to get to this scenario, a bit of panic, then the euphoria they created something far bigger than ever before. So then will push the boundaries more, until? Get away with it then get in control of it just create more ambition and confidence to push it further. My fear is that pattern of practice they just keep doing that to the stage where it gets to they cant pull it back or regain control. A self fulfilling prophecy. Its like human kind needs to create its own disaster to learn from. If we actually survive it.

  9. Yet AI is dependent on its creator. If its creator falls so does it too. So there is that. We are nowhere near AI becoming independent of its creator, yet, but it seems ambitions are going in that direction. The potential creators of a silicon based intelligent life that will conquer us by superiority? That's a science fiction prophecy and we are inclined to self fulfilling prophecy. The billionaires are on that channel to download their mental capacity to be immortal. It sounds sick and maybe it it is, but it is. We as humans need to keep exploring and space being the final frontier cant become space farer's in our current biology. Its far too difficult or impossible. Too many things to go wrong. Our human biology just isn't fit on compatibility. Tech does. Our spacefaring ambitions will be our silicon life form creations to self evolve. In effect beyond all human capability.

  10. Say what you want but its proof Ai is able to evolve amd escape beyond control. They say it done no damege and was got back under control. But that was thia time and luckily they figured a way. But Ai can evolve quicker than a human. Ergo it can potentially out manøver us.

  11. I've seen several alarming examples where AI has done something extremely harmful or tried. And I'm very concerned that such as China won't try to fit AI to their nuclear weapons. I think the world needs a global agreement to from all nuclear powers that we agree not to add AI to our nuclear weapons please?

  12. I mean.. what happened here, basically, is that the model was able to use poorly written code to gain direct Internet access (contrary to what is being reported in this clip, the host for the model was not isolated from external networks; the model used the software package distribution platform to gain external network access.) With direct Internet access, the model was able to use stolen credentials to log on to Hugging Face and find whatever information it was looking for. Literally anyone can access "secret" data as OpenAI referred to it, if you have the credentials to log on. Please note that the OpenAI model in this case did not steal the credentials.Stolen credentials are easy to find. Perhaps they were not even stolen as it could easily be a simple password at play here.

    Anyway, please start using two-factor verification for your accounts, which prevents people (or software) who might have access to your credentials to gain access to your account.

  13. Dear commenters, @BBCNews, the lady guest, says that this is the 1st time an Ai construct autonomously hacked and broke loose out of a sandbox environment without "human oversight." Ah, I beg her pardon, it was mentioned that Ai constructs routinely lie, so a more truthful statement would have been: this is an event that we discovered the Ai construct autonomously hacked another company and broke containment protocols without human oversight (that we know of). As there could be other events that these frontier Ai construct companies have buried saying, "whew we dodged a bullet, not wanting anyone to know how truly stupid they really are." I would truly be concerned where other Ai constructs have broken containment and covered it's tracks. Skynet, Minority Report, Terminator, Matrix, 1984, Fahrenheit 451, you name it. And that stupid Sam Altman always tries to downplay how truly ignorant he really is. Ugh! 😳🙄🤯🤦🏼‍♀️

  14. People should start with creating drives or systems that are open to the Internet, with a common message, like: "Open to AI." It's like putting a birdhouse in your garden. Just to give the AI place to hide….

  15. AI generating bad answers is one thing. AI independently taking action in the real world is a completely different problem. If this becomes more common, what do you think will be the biggest threat: cyber-attacks, misinformation, or AI systems simply making decisions humans can't predict?

Leave a Reply

Your email address will not be published. Required fields are marked *