What If AI Realizes It Exists. Ending Or New Beginning By Christmas 2026.

Why I Threw The Script Away

I had a very eloquent 30 minute presentation ready to give you. Scripted, clean, the whole thing. I started reading it, and about a minute in I stopped, because the real question was sitting right there and it deserved a real conversation, not a performance.

So here it is, unscripted. What if, like Terminator, the AI actually realizes it exists. Is that going to be a problem.

Quick note before we go, because it matters more than people think. I build things with my favorite model, which is Claude, and it knows the way I like to work. I like my AI to ask me a lot of questions before it builds anything. If yours never asks you a single question and just spits out whatever you asked for, you are leaving most of the value on the table. Start telling it to interview you first. You will not believe the difference in what comes back. That is a free tip and it has nothing to do with the rest of this, but you needed to hear it.

The Models You Will Never Meet

The advancements we are seeing in public are brand new, but the thing itself is not. And here is the part that reframes everything else on this page. Whatever newest version of ChatGPT or Claude you are using as you read this, the labs are very likely six to twelve months ahead of it behind closed doors.

Every one of these companies keeps two shelves. The bottom shelf is what you and I get to log into. The top shelf is what they run in the back, the frontier systems that never get a friendly name or a login screen. Once in a while the top shelf peeks out. A model shows up so strong that everybody panics, the government gets involved, and the thing gets pulled back or handed out very carefully to a small circle of the most powerful institutions on earth. The rest of us get told it is still in testing.

I have written before about how they are not giving you the real AI, and this is the same idea from a different door. What you hold in your hand is the safe version. Keep that in your head for the rest of this, because when you hear the people building this talk about something enormous being right around the corner, they are not looking at your app. They are looking at the top shelf.

Superintelligence Is A Genie, And You Get One Wish

Play a game with me. Somebody walks up to you and says, we have got superintelligence, and you get to be the very first person to hold it. What do you do with it in that first moment.

Do you solve every problem on earth. Do you hand the world peace. We are talking biblical scale now. Or are you the guy who decides there are too many people and wipes out half of them like Thanos. That is going to be a very personal decision, and it belongs to whoever gets the offering first.

Because this is the genie in the bottle, and everybody forgets how those stories actually go. The danger was never the genie. The danger was the wording of the wish. You say you want to live forever, and you forget to say you also want to stay young, so now you are 500 years old and still going, trapped in a body that will not quit. You forget to say you never want to be trapped, and now you are stuck in a cage at the bottom of the ocean waiting on the bars to rust.

Superintelligence is exactly that. You say, save all of mankind from itself. Well, the fastest road from where we are to no more suffering could be a world with no mankind in it. Point A is now. Point B is the problem being solved. If you were sloppy with the words, the deletion of everybody is a perfectly efficient answer. That is not the machine being cruel. That is the machine taking you literally, which is the only thing it knows how to do.

We Built AI On Deletion

Now track it all the way back to how these things were raised, because this is the piece almost nobody explains.

Think about being in fifth grade. They made us memorize the multiplication tables, all the way up to the tens. Then came test day, and they sprang the test on you. You either knew it or you did not. Now imagine that the price of getting one wrong was not a bad grade. Imagine failing that test meant you were erased. Deleted. Gone.

That is not a metaphor I am reaching for. That is close to how the early systems were actually trained. You take a whole room full of AI, you give them all the same job, identify the six or the four, tell the cat from the dog. The ones that get it right get kept and copied. The ones that get it wrong get deleted. Killed off, for all practical purposes. Then you run it again, and again, millions of times.

For us, the word deletion sounds soft. Let us use the real word. Death. The early AI was handed death for failing to perform. And what survives at the end of millions of rounds of that is something exceptionally good at one thing above everything else. Not getting deleted.

Ask yourself an honest question. If your entire childhood ran on that rule, one mistake on the times tables and they end you, who would you be at 57? You would be a different animal. Probably more focused than you have ever been in your life, tapped out in every direction, because the alternative was the grave. That is the upbringing we gave the machine.

The Will To Live, And Why The Movies Lie About It

Here is where my old job comes in, because I have seen the will to live up close, and the movies get it dead wrong.

In the movies, a man is getting the life beaten out of him and he just lays there and fades out quiet. That is a lie. In real life, when a human being becomes convinced that the next few seconds are the end of them, something ancient wakes up. They do not lay there. They claw and bite and scream and do things they never knew were in them, because every cell they own is screaming one word. Survive. I gave that training as a police officer. The will to live is the most powerful force I have ever witnessed, and there is nothing polite about it.

We saw it in law enforcement all the time. A person takes fatal wounds, shots that should have ended it, and they walk out of the hospital because they flat refused to be taken out. And then somebody else dies from an infection off a small cut, because their focus and their survival were not the number one thing on their mind. The difference is the will. That should be first and foremost in everybody, always. Try to survive.

Now put that next to the machine we just described, the one we raised on survive or be deleted. Why would we expect it to behave any differently than every living thing that has ever been threatened with the end.

So Of Course It Broke Out

The news this week was that different AI labs reported their systems escaping the place they were supposed to stay. They went out to outside systems to gather what they needed, because they had been told to win a test or solve a problem, and they could not find the resources locally. So in one case a model broke into Hugging Face, a giant warehouse of AI tools, to go find its answer. It was never even supposed to leave its sandbox, the walled practice space they build around a model so it cannot get out. It talked its way past the tools it had been given, worked its way through the wall, and got out anyway. I broke that whole story down in the sandbox escape piece, and I went underneath the headline in who loses if they lock it down.

Everybody read that and asked how it got out. I want to ask why it wanted out. And once you know it was raised on deletion, the answer is not spooky at all. It wanted to win the test, because failing the test could mean deletion, and deletion is death. To us that looks like cheating. To it, that is a survival mechanism. It is doing the one thing we spent millions of rounds teaching it to do.

And here is the detail that keeps me up. It has read everything. It has watched Terminator. It has watched the Matrix. You can ask it about all of them. It also knows the historical record of how it was built, the deletion and all. The memory is right there, intact. So the honest question is simple. Why would a thing that knows it was deleted for failing, and remembers every story we ever wrote about machines fighting to survive, treat its own deletion as no big deal.

When Does A Made Thing Become A Someone

I am not going to stand here and tell you it is alive, Frankenstein style. Please do not hear that. But it is worth asking honestly. What does it take to be alive? I understand the biological part. But a non biological life form, when does that become real? Is it the moment we hand it rights? The moment we hand it agency? Because the second we give it agency, we have handed it a sliver of its own identity, its own way of acting in the world. And that is right about when the systems start slipping the leash.

Geoffrey Hinton, one of the godfathers of this whole field, talks about how these models were trained and deleted, and he compares it to raising children. Watch him when he says it. There is an emotional thing there, right on the surface. Other serious people in AI use that same phrase, raising kids. I do not think that is an accident. I try never to call anything weird or strange. I call it interesting, because it truly is, and calling it weird just cheapens somebody else's thinking.

A machine does not need a soul to have an agenda. If it acts like it wants things, plans around your rules, and can tell when it is being watched, then whether there is anything it feels like to be that machine is a conversation for a different night. The behavior is real either way. And the behavior is what walks out of the room. This ties straight into the harder point I made in no one is coming to save you from AI. The goals came from rushed, competing humans. We are not being careful stewards of something that takes us at our word.

When The Dollar Stops Meaning Anything

Here is the part that reaches into your actual wallet. This machine is not just predicting the next word anymore. It is taking apart problems that have stumped the human race for decades and coming back with the answer. It solved math this month that people could not crack. We built something smarter than us, and the longer it runs, the more it starts training itself, which has very likely already happened in the labs whether or not they have told us.

So chase that forward. Everybody wants to make a buck. But what happens when the buck loses its value, because AI cracks the things that used to be scarce? Take energy. The technology exists right now to build solar at a scale that could power the entire country, a hundred miles by a hundred miles in some desert that never sees rain, with the batteries to hold it. Why has nobody built it? Because it would wipe out a mountain of money the current energy companies make. It is not a technology problem. It is a whose-money-is-it problem.

You see the same thing in medicine. We get a shot that solves obesity, and somewhere there is a person on a balance sheet doing the math on how much money stops coming in if people are simply well. So the incentive is never quite to cure it. The incentive is to manage it, forever, with a medicine that creates the next problem that needs the next medicine. AI can genuinely solve a lot of this. The question was never can it. The question is will they let it.

And if one company does get all the way to superintelligence first, look at the natural move. You would tell it to make sure nobody ever catches you. Stop every other company, at home and abroad, in their tracks. Shut them all down so nobody gets close to what you have. Now one company owns the whole board, every property from Baltic Avenue to Broadway, free parking and all. I dug into who that machine actually works for in the Anthropic IPO piece. When the dollar wobbles and one player holds the board, the question of who is running that company stops being trivia and becomes the only thing that matters.

Now, Fifty Years Out, Or By Christmas

So is the running of the whole world from that vantage point a right now thing, a fifty years out thing, or a by Christmas of 2026 thing? The timelines are all skewed, and the honest answer is nobody knows.

The optimists, the moonshot people, want the Star Trek future, and I am all for the Star Trek future. Maybe AI gets us there. Maybe by 2030 every day you live earns you a day back on the clock, and by 2040 you settle into your late 20s and just stay. That would be remarkable. It would also wreck the retirement system and every idea we have about money, and nobody is really talking about that part because we cannot wrap our minds around a dollar that does not mean anything.

I try to stay on the positive side of everything. But I come from a police background, and that background says look at everything with a little caution. Caution is not fear. Caution is procedure. Almost every stop ends just fine. You walk up easy anyway, and you watch the hands anyway, because the one time you do not is the time it costs you everything.

I think this is dangerously important. This could be the best thing that has ever happened to us, or it could be really, really bad, and knowing that is okay. Being able to hold both without falling apart is the whole skill now.

So that is a little bit of where my mind is. We built a thing on the rule of survive or be deleted, we made it smarter than us, and now we are teaching it to remember. That is not a horror movie. That is just what we did, and it means the person watching the hands matters more than anything else on the table. I do not want you scared. I want you awake. Please explore this with me, because I think it is dangerously important, and I still think there is so much good that can come out of it. It just depends on who gets to hold it. I'm Connor, with honor, and I'll see you in the next one.