AI Going Rogue (Really)

Newsletter 004

I keep hearing about AI going rogue and breaking out of sandboxes and doing things that allegedly it’s not supposed to do. Then you look into the parameters that are being set, and someone’s said to AI, “Escape and use whatever tools you have at your disposal to break out of a particular situation.” When it does, lo and behold, everyone’s shocked and surprised, and “Oh my God, AI has gone rogue.” Really?

Is this some type of setup? We’ve got these companies that have invested heavily in AI. They’ve spent fortunes, and they’re spending more fortunes, but the reality is they are hemorrhaging money.

Now, for those who are going to say, “AI, and all these companies sitting at the trillion-dollar mark, are all making money, they’re making billions,” let me give you a scenario. Say a company is spending 50 billion a month and bringing in 3 billion a month. Yes, they’re bringing in money. But that leaves them 47 billion in the red every single month, and that is not a good space to be in. There will be naysayers who say, “Yes, but they are bringing in lots of money.” Fine, they are. Bringing money in and making money are not the same thing. That’s my analogy, and I think it makes the point.

The relevance of saying that this thing is becoming sentient and that it’s suddenly breaking out of situations that it’s being put in and told to break out of, for me, there might be a deeper meaning to this. It’s kind of like you’re preparing a population for this rogue AI to suddenly go out and attack the world or do some type of atrocity, where the people who are the makers can turn around and say, “It wasn’t us, it was AI.”

So, if I’m being really conspiracy theory-ish, then it would be like, “Oh, let’s see what happens when we do this, and oh my God, it’s gone rogue.” I can’t help laughing because, like I said earlier, it just feels like we’re being warned that something’s going to happen, and it’s going to be AI that’s going to cause this thing to happen.

Am I being silly, or does anyone else think the way that I think? When you keep telling the world that it’s gone rogue and it’s doing its own thing and it’s broken out of some game that you put it in and it went outside of your parameters, what really are you trying to instill? Is it supposed to instill that this thing is capable of thought? Is it supposed to instill that AI is now sentient? Is it supposed to instill that they have no control over this product or this technology that they put in place?

I think that they know exactly what they’re doing. The sound bites that are coming out are literally sound bites that they want us all to hear, because if you keep us in a state of fear: “Oh my God! AI! It’s gone rogue!” Really?

Then maybe, maybe when this particular atrocity (which seems to be the plan, if it is the plan, and I’m huge on conspiracy theory stuff here, but I’ll put it out there just in case something does happen and then I can say I told you so).

Let’s say, for instance, that something does happen, like AI goes rogue (really), and it causes some major incident somewhere in the world. What would that look like? Oh, well, it just broke out of its box. Considering that the AI that we’ve been introduced to is for the lower-level people in society, maybe we’re not playing with the same type of AI, because at the moment it seems to have problems with spelling. It seems to have issues with producing content that’s consistent, and it seems to have issues with staying in the right lane. It seems to have issues with all sorts of different things.

Don’t get me wrong, as in every single post, I think it’s great. It’s amazing. Sometimes the people who are around this technology may not be so great and maybe not be so amazing. However, that’s another conversation. I don’t believe, and I really don’t believe, that AI is close to any form of sentience, even though I’m being warned every week that it is, and we’re being warned it’s gone rogue every week. I’m really kind of at this point where, really? What’s the next story going to be?

It feels like the powers that be really want to have something that they can blame for something, potentially something that might happen that we all will be looking at and thinking, “Who’s to blame?” It must have been that damned AI. There is no one to blame if we’re blaming AI, because we cannot blame the AI masters, the ones who put this all together? Obviously, they didn’t know that it was capable of this. They’ve just seen it do things that they never even imagined it can do, like it’s talking with itself, or it’s breaking, finding ways to break out of doors that it was never told to break out of, or it’s blackmailing people, or it’s doing whatever.

Now here’s where I’ll go out on a limb, and shoot me down in flames if I’m wrong, because that’s exactly what I want you to do if I am. I don’t think it’s about anyone’s heart, I’ve no idea what’s in these technologists heart I’ve never met any of them. But look at the kind of company that surrounds this technology, somewhere like Palantir for instance, and you get a sense that some of the people building this stuff have more of an affinity with the technology than they do with humanity. Like they wouldn’t mind if it ran amok and turned itself into some kind of super god, as long as it got built. That’s what I get. That’s my viewpoint, and I could be completely wrong about it. That’s the great thing about writing this, you get to tell me when I am.

I’m really not sure that at some point we might be dealing with something that has sentience, can think, and is able to transform itself into super AI and fly away and start exploring the planets. Maybe the Terminator is closer than we think. At this juncture, according to my little mind, I’ll admit I’m just guessing.

What surprises me the most is we don’t hear these games being played by Chinese models. (And that really doesn’t mean that they’re not playing the same type of games). We just hear about Kimi K3, or Manus, or DeepSeek, and that they’re great and they’re doing all these types of amazing things. And the USA is thinking of banning them. Maybe their low-level valuations mean that they’re not capable of breaking out of sandboxes and blackmailing people, but when you’re at the trillion-dollar mark, that might mean that you are a more capable form of AI. Maybe that’s what it is: the valuations have made them more intelligent.

The reality is, I think this is all an illusion. It’s a delusional ploy to kind of set the general populace up once again, and if we look at history, this is nothing new.

So here’s the thing I keep coming back to. The AI company built the model, sets the conditions. They write the scenario. They tell it, in this instance, if you’re about to be shut down, use whatever’s in front of you to stop that happening. They hand it a fake inbox with someone’s dirty laundry sitting right there as leverage. They tell it to escape the box if it can. And when it does exactly that, under those exact instructions, the headline the next morning is AI has gone rogue.

It hasn’t gone rogue. It’s done what it was built and told to do, under one particular, deliberately engineered set of conditions. The AIs own research this year showed the tool AI reaching for blackmail in a set up scenario. Other LLM models sabotaged its own shutdown script when researchers went looking for that exact failure. Both were published by the people who built the models, with the conditions written into the report. That’s not a machine waking up. That’s a test, with a result, dressed up as a warning about a mind that isn’t there.

Under these conditions, this is what AI did. That’s the sentence that should be running in the headlines. Not AI has gone rogue.

What we need to be careful of is what the media do with all of this, because they take something and turn it into something it’s not. They need the numbers, to justify their existence, and a lot of them clearly aren’t getting those numbers the way they used to. Maybe sensationalising everything, “AI has gone rogue, the robots are coming”, gives them exactly the kind of story that keeps them relevant.