They Built AI Too Powerful For You To Have
June 27, 2026 // Daily Download // Connor MacIvorWhy is everybody so irritated at the AI labs and the government, at what looks like overreach? Maybe it is not overreach. Maybe it is. But this is what is actually happening in the world with artificial intelligence right now, the end of June 2026, and it is worth looking at straight. The pattern is simple. You have it one moment, and then you do not.
The Model They Locked Away
Take Anthropic, the company behind Claude. You have probably heard of Claude by now. A lot of people love it, a lot of people use it, and it has parts and pieces under the hood that power real work. A few months back they put out a particular model that became incredibly popular, and incredibly powerful. People went crazy for it. And then the story shifted. It was so unbelievably powerful, the framing went, that regular people could not be trusted with it. Too dangerous. So they put it under lock and key.
But here is the part that sticks in the throat. They did not just shut it off. They kept it available for the people running the financial world, the Forbes-list crowd. So the most powerful tool gets pulled away from the public for safety, and handed to the wealthiest people on the planet at the same time. Of course people got upset. Tell someone they cannot have something, and watch what it does to them. They could not have it, so they were not happy.
Snatched Back In Three Days
So then, maybe to smooth it over, a less powerful model gets let out to the public. A weaker tier. And here is where I can tell you from the inside of the user seat. I keep the high-end account because of the things I build and the work I do with AI, which is also why I can sit here and try to help you understand what is going on. I saw the button. I could have clicked it and activated the more capable option. And it would have lasted me maybe three days, because after a few days on the market they snatched it back, citing too much of a security concern. So the public does not get that model. The pattern again. Here, then gone.
Real Danger Or Theater
Now, is all of this true, or is it clever advertising and marketing? Honestly, it is hard to say. If you look at track records, some of these large language model companies are all about drama. Oh my gosh, watch out, this is going to be the thing that changes the world. Does that sell newspapers? You bet it does. But maybe it is also true. I am not on the inside. The thing I think people are not chewing on long enough is this. If we are seeing these models, and the labs are talking about releasing them while telling us how dangerous they are, then what exactly are they working with? What do they have access to that we never see?
These are private organizations, not government entities. And I am not going to pretend the government is some perfectly safe steward either. Track records on that side are not spotless. But there is an argument that some level of control keeps us from doing real damage to ourselves, the way certain things are kept illegal so people do not hurt themselves and create a mess for everyone. Maybe with something this big we do need a measure of that overreach, because what we are dealing with is intelligence. Intelligence beyond what we have ever conceived. The people at the top of these platforms, the partners with billions in the game, they say it themselves. I dug into who actually wins from all this when I asked what the endgame of superintelligence really is, and into the safety drama itself in the sermon against AI safety theater.
Why Code Broke It Open
Right now these systems are starting to learn through recursive self-improvement. They look at their own code, they rewrite it, they teach themselves new things, and they better themselves by writing new code. And code, it turns out, is the thing that broke the whole industry open about a year, year and a half ago. Not images. Not video. Code is the most important output, at least for now. The ChatGPTs of the world, Claude, Google Gemini, even the Chinese models that are sort of open source but not all the way, they are all very good at producing code. And what do you do with code? You build things. I broke down what it means when a machine built a smarter machine without us, and why a system copying another model is its own story in the post on how China copied an AI by just asking it.
You can say, build me an email inbox that vets every message and responds a certain way. Build me a contact manager so I can track everyone and parse my email every single day. Wire in a calendar. Build a tool that finds the best grocery prices, or one that does the heavy lifting of planning a trip. There are a thousand things these systems can do. And when people have access, they can do genuinely wonderful things. Curing disease. Fixing the fat problem. Working on world hunger, on the economy, even on the grinding human conflicts that never seem to end. The capacity is real. We just spend most of our energy arguing instead of pointing it at the things that matter.
The Jobs That Actually Get Replaced
Here is where I part ways with the official line. The people running the show are very firm that AI will never replace humans. They do not even want you to say it. AI is AI, they say, and it will never replace people. I think they are missing what most of us actually live with. Let me use myself. I am a real estate agent, and I do other things. Could the real estate part be replicated, cloned, and run end to end by AI without a human overseeing it? Sure. Probably. And then take something fancy, brain surgery. Could a robot, maybe not human-shaped, do that surgery? Some of that is already being done. The open question is whether it runs the entire arc, from the first consultation and the options to booking the surgery, doing it, and managing the recovery, with maybe one human nominally overseeing it, limited mostly by lawsuit concerns.
We already see it in radiology. Machines read scans and call what they see more reliably than people do, missing less than humans miss. A doctor working alone lands somewhere around 70 to 80 percent diagnostic accuracy. Put AI on top of a human doctor and it only nudges up a little. But take the human out and let AI run the whole thing, and it gets substantially better. At some point the insurance companies do the math and decide they cannot afford a human in the loop, because when the human makes the mistake, the lawsuit lands on them. I traced this same labor logic when AI replaces your workers and we asked who is left to buy the product.
Now bring that to cars. There is going to come a point, probably soon, where it is more of a liability to have a human operating the vehicle. We saw a version of this in the Will Smith movie, where you could grab the wheel and drive yourself, but it was strongly discouraged. More than likely the insurance companies get there. Drive it yourself, and the premium is enormous. Let the car drive you, and the insurance is practically free, because the machine does not make the mistakes. Stolen cars get rare too, because we are all tracked through the nose at that point. Privacy is already mostly gone. We probably signed it away ourselves, in some terms of service we did not read when we got the iPhone or the smart TV. Who really reads the small print?
Artificial Is A Terrible Name
Then you have the ladder everyone keeps pointing at. Narrow tools first, the ones that are great at a single game like chess or Go. Then AGI, artificial general intelligence, a system that does well across the board instead of at one task. Then superintelligence, an entirely different level above that. Point a capable model at cancer research and have it work the problem. Point it at longevity, at the economy, at starvation, at the planet. The capability to chip at those is arguably here already, and yet we are still busy arguing over the Strait of Hormuz, over Iran, over why people cannot get along. Argentina is running an experiment where AI oversees a lot of the machinery of governance, which is fascinating and also a little terrifying if it spreads and then stumbles.
And I will say this plainly. Artificial is a horrible name for it. When I hear artificial I think fake, lesser, not as good as the original. But the original intelligence was us. We are the ones who assembled this thing, got it to start growing, and then watched it take on a life of its own. There is nothing fake about what it does. It is real intelligence, just built rather than born. That being the case, it is probably too late to shut it off. It is moving too fast. I went deep on whether any of it is truly aware, or just good enough to fool us, in the post on AI consciousness. Honestly, when it gets convincing enough, we will not be able to tell the difference, and to us it will be the same either way.
The Genie Warning
Here is the picture that keeps me up. Pretty soon these systems may tell you to do something that makes no sense in the moment. Connor, tomorrow get out of the right side of the bed instead of the left. Three hours later you will have an idea. Write it on a cocktail napkin and bring it back to me. I am not sure what it is yet, but it all connects, and from there we will get you the health insurance you have been after. And you think, that is silly, I always roll out the left side and ignore it. But maybe that move really is connected to something the system can see and you cannot. That is the intelligence we are building, or rather the intelligence the entities we created are becoming. I do not even know what that looks like, but that is the potential.
And maybe it is good that humans are still driving it, because of the genie problem. What if it is one of those genies that takes everything completely literally? You ask for a thing, and it gives you exactly that thing, but you did not think it through. You ask to be eternal, but you forget the fine print, the additional wishes. So you end up buried in a box, reinforced concrete all around you, eternal, unable to end it, unable to starve, stuck there forever because you thought you outsmarted the genie. That is the care and consideration this deserves. The power is real, and so is the cost of asking for the wrong thing carelessly. It is the same caution I raised when the government asked an AI lab to slow down.
And honestly, a lot of what we are watching at the top looks like wrestling. The match is real, sure, but there is a ton of choreography behind it. When somebody tells you they have something so powerful they cannot give it to you, and then they finally hand it over, you get more excited about it and you value it more than if it had just been sitting there. Tell people they cannot do something, and what do they want to do? That thing. The labs are doing a dance. They have their fans, their investors, the early money buying in before any of us regular people can touch it. And they have done very well doing it.
You Do Not Have To Be An Entrepreneur
The last thing, because everybody keeps saying it. You are going to have to become an artist. Find your entrepreneurial lever and flip it. Use AI to become the entrepreneur you were always meant to be. And I do not know if everybody is built for that, or even wants it. Some of us complain about the day job and are still kind of happy with it. Being a cop is hard and underpaid and grinding, and there is also a real good that comes with doing it well, especially when you are one of the good ones. Plenty of people do not want to be entrepreneurs. They want enough money to not live in a panic. The people loudest about the entrepreneur path are usually already at the level where money is not the daily worry, and they are selling that path. So learn the tools, get genuinely good with them, make yourself harder to replace and more valuable wherever you already are. That is the move that works whether or not you ever start a thing.
Selling in Santa Clarita?
I represent sellers only, 100 percent on your side, no dual-agency conflict. One flat fee of 17,000 dollars to sell your home, with every other cost that touches your equity examined and negotiated. The same work whether your home is 500,000 dollars or 1.5 million dollars, so the same fee.
Sellers Only AgentThe Daily Download in Your Inbox
Weekday morning AI news, weekly SCV real estate market deep-dives, and long-form essays when something needs to be said. No spam. Unsubscribe anytime.
FAQ
Why do AI companies release a powerful model and then take it back?
The stated reason is safety. A model gets called too powerful or too much of a security concern, so it is pulled from the public and access is kept for enterprise and high-end accounts. Whether that is genuine caution or a marketing move is the real question. Telling people they cannot have something makes them want it more, and a model that is rationed and hyped feels more valuable when they finally get a weaker version of it. From the outside it is hard to separate true risk from theater, but the pattern of release, hype, then pullback repeats often enough to be worth questioning.
What is recursive self-improvement in AI?
Recursive self-improvement is when an AI system works on its own code, rewrites parts of itself, and uses those improvements to get better, then repeats the loop. The labs building these systems say it is already happening. The concern is speed and visibility. If a system is improving itself faster than people can review what it changed, the humans running it may not fully understand what it is doing moment to moment. Code turned out to be the thing these models are best at producing, which is exactly why a model that can improve its own code is a different kind of tool than one that just writes text.
Will AI actually replace humans in jobs like surgery, radiology, and driving?
In narrow tasks it already beats human accuracy in places. AI reads certain medical scans more reliably than human radiologists on average, and a doctor's unassisted diagnostic accuracy sits well below what a fully AI-driven process can reach in some studies. The push toward replacement may come less from the technology and more from the insurance math. If a machine makes fewer mistakes, insurers will eventually treat the human in the loop as the liability, not the safeguard. The same logic applies to driving, where an AI that reacts faster and crashes less could make human-driven cars the expensive, discouraged option.
What is the difference between AGI and superintelligence?
Today's models are narrow. One is great at a game like chess or Go, another is great at language, but they are pointed at specific tasks. Artificial general intelligence, or AGI, would be a single system that performs well across the board, the way a person can move between very different problems. Superintelligence is the level beyond that, a system smarter than humans across nearly everything, capable of producing things we may not even be able to evaluate because they are beyond our ability to comprehend. The word artificial is misleading, because what these systems do is real intelligence, just built rather than born.
Do you have to become an entrepreneur to survive the AI shift?
No. A lot of advice says everyone has to find their inner entrepreneur and use AI to build a business, but not everyone wants that or is built for it, and that is fine. Plenty of people are genuinely satisfied doing good work at a job they care about and simply want enough income to not live in a panic. The honest move is to learn the tools so you are harder to replace and more valuable wherever you are, rather than assuming the only safe path is to go start something. The people loudest about the entrepreneur path are often already wealthy and selling that path.
That is where we are, June 27, 2026. The labs are doing their dance, the most powerful tools keep getting dangled and pulled back, and the rest of us are trying to read the small print before it reads us. I am not a billionaire and I am not a doomer. I am a working guy who builds with this every day and wants you to see it clearly. Learn the tools, stay connected to the humans, and be careful what you wish for. I'm Connor with honor, and I'll see you in the next one.