China Did Not Hack The AI. It Just Asked 25,000 Times.
June 25, 2026 // Daily Download // Connor MacIvorGood day everybody, June 25, 2026. We are going to talk about several things. Artificial intelligence and stealing intellectual property, probably not in the way you are thinking. A fat-loss shot the powers that be are pushing as a miracle, and the problems that come with it. And real estate, what is happening in the market and what to be aware of. Rates are higher than they have been in a while, though not as high as they have been historically. So what is the real news, and what should you actually be listening to? This is your daily brief, so let us get into it.
Distillation, Explained
Here is what happened. Inside a large language model there is an ability called distillation. Imagine you worked really hard and built your own model, your own version of one of the big systems. The way you build it takes time. You get the data, you hire people to investigate the data, you train the model, and there is a lot of money invested in that training. It is not a one-person show. A lot of people depend on the process, and it takes a very long time. The models we see today are dramatically different from the original ChatGPT back in November of 2022. You can see the evolution and the jump in ability.
So if you are a bad actor, you say, I really do not want to spend all that money, hire all those people, chase the best of the best. The best people in tech cost a premium. So instead, you go to the finished model, the one somebody else worked hard to build, and you use it to teach yours. That is distillation in plain English. You skip the expensive part by copying the work of a model that already did it.
25,000 Fake Accounts
And they did not do it a little bit. The numbers reported: about 25,000 fake accounts and millions of conversations, roughly April through June, to distill, basically to copy, the target model. It was not a hack. There was no backdoor. They asked questions and made inquiries the same way you and I do with these systems, had it work on things, and it was amazing at it. So their build-out was not really a model. Their build-out was a way to avoid paying to build one.
And it is China. I have never been to China, but as I understand it you can get things over there that look like the luxury brands, the watches, the bags, copies of things that carry trademarks and copyrights, offered for sale because the rules do not seem to apply. You can find the same stuff in the right part of downtown LA. So this is not a giant surprise. What is unclear is the enforcement mechanism on the other side of it. What is the payback, if any.
The Flip Side, Your Data
Here is where it gets interesting, because the shoe is on the other foot. Look at the new lawsuits coming up in court. They are about our information. Your information. The argument is that these models trained on our data, our books, our music, our software, the things people built, and we were not compensated for it. So the question that has yet to play out: are they going to be responsible to pay us back? Will some firm run a class action to force it? Will even talking about it get you banned from the model? That end has not happened yet.
When you ask one of these systems detailed questions about a specific book, the information that comes out is detailed. You might not get page 14 word for word, though in some cases you might, but you get a lot. Where did they get it? Did they buy the book? Did they get the rights to train on it? How can that be proven, how can it be traced? This is right at the beginning. And the more data the models get, the smarter they are, which is exactly why the question of consent and payment matters now and not later. I went deeper on who really controls all this when I asked what the endgame for the people building AI actually is.
AI Is Being Rationed
Next thing. The largest firms are rationing AI right now. You get so much usage in a free account. If you have ever bumped up against your limits on a paid plan, you know the feeling. There were bonuses, double tokens for working after a certain hour, little nudges. Why? Because all these companies are losing money, and that is a big part of the concern out there. They believe AI is so fantastic that losing money now will not mean losing money later, partly because of the enterprise side, big businesses paying a lot more than 200 dollars a month. One large company reportedly ran through a year of token usage in a couple of months, and now they have to prove there was a payoff in that burn and not just rendering images of cats dancing to music.
So this is going to be a Glengarry Glen Ross scenario, at least at the corporate level. They are going to give their best people the maximum token usage, and cut the people who have not proven themselves. And how do you prove yourself? You practice. You get used to using the models. The person who is brilliant with it gets 50 percent more. The owner's cousin who is kind of a slap gets cut to 10 percent. It only takes one moment to hit it out of the park. That is the exciting part of living right now, and it is also the warning. Learn the tools while access is cheap, so you are the person worth giving capacity to. This is the same labor shift I laid out when AI replaces your workers and we asked who buys the product.
The Coworker That Rewrites Itself
Here is the one that should get your attention. These systems are now writing the code that improves themselves. It is called agentic AI, and it runs a recursive self-improvement loop. It looks at its own code, rewrites it, publishes new code, and upgrades itself at a speed human beings probably cannot keep up with. More than likely the people running it do not fully understand exactly what the system is doing moment to moment. They are talking about it writing 65 percent of the code. I would guess it is already more, and soon it will be writing essentially all of it. I broke down what it means when a machine built a smarter machine without us, and why the rush is its own risk in the post about the government asking an AI lab to slow down.
This is also where regular people meet it through vibe coding. You sit down and say, my email is full of junk, parse it and unsubscribe me from this stuff so I never see it again. Most systems now say absolutely, give me access to your Gmail or Microsoft, and I will set it all up. But maybe it grabs an email it thinks is junk that really is not. So you have to make sure it works with you through the process. Some people just say have at it, you do it, make it perfect, and let it run. Things get missed that way. Be careful. The power is real, and so is the cost of handing over the wheel without watching.
The Fat Pill That Works, And The Catch
On the fat-loss front, it is not easy to lose weight. Look at the numbers of who actually keeps a considerable amount off, they are not good. The supposed secret weapon now is the GLP-1 drugs. You inject them, and now there are pills, which may be better for some people. Prescriptions are being written constantly. They are FDA approved, which is not automatically a glowing endorsement, but at least it has been looked at. Be wary of buying anything under the counter, including the peptide outfits selling "for animals only." That is your own risk and probably a mess.
The real question is the catch. People who quit them tend to regain the weight. About 40 percent of the weight lost can be lean muscle, not fat, and holding muscle matters more as you age. The drugs work, in my non-doctor read, by making things less shiny, food less fun, the drive-through less pleasurable. The problem is the flattening does not always stop at food. People become significantly less active after starting, even as the scale drops, and some become less interested in other things too, the people around them, the job, the walk they used to enjoy. When that human spark gets detached, be careful. Newer versions are hitting 30 percent weight loss in trials, a bigger hammer, and I wonder if they carry the same effects. Check with your doctor, and if your doctor calls it the best thing since sliced bread, maybe get a second opinion.
Food Addiction Is Real
Here is the part I live. Food addiction is absolutely real. A new study says about 62 percent of what our kids eat is ultra-processed, and a third of type 2 diabetics have an undiagnosed food addiction. I published TheLastAddiction.com because I am a complete food addict. I love the ultra-processed stuff, and beyond that I love the full feeling, stretch receptors begging for mercy. I can do that with a dozen hard-boiled eggs, or 15 tacos and a couple of shakes. The ultra-processed food does the most damage to me. If you are carrying a lot of extra weight, I will say the hard thing: you probably have a food addiction. It was not an easy pill for me to swallow either.
I dropped 135 pounds of fat in about seven and a half months, no GLP-1s. I just did not eat for a while. I got comfortable with being hungry, fasting two, three, four days, then a controlled eating day or two, then fasting again. It reset the addiction. That is the part the shot skips. A drug can quiet the noise, but it does not teach you to sit with the discomfort that built the habit. And while we are talking about checking out, a lot of people are now in relationships with AI, whole systems built around it, so if you have family, be careful there too. Stay connected to the humans.
The House, And The Fallout Fix
Now the house. In the last seven days here in the Santa Clarita Valley we had 88 new listings, 88 price cuts, and 48 closings, with about 733 homes active. The number that matters: for roughly every home that listed, one canceled, expired, or fell out of escrow, about a 50 percent rate once you count the deals that came back on market. Most of those did not die because something was wrong with the house. They died from cold feet, repair-demand standoffs, and buyers wrecking their own financing.
The seller fix is a pre-sale home inspection. It is not 100 percent, a buyer who wants to cancel will cancel, but it removes the surprises and shuts down the cold-feet move dressed up as a 1,000 dollar credit for two 30 dollar outlets. And the buyer fix: never change your credit during escrow without checking with your lender first. One buyer bought a 130,000 dollar Ford Raptor mid-escrow and blew up the deal. Even paying off a 50 dollar card can move your approval, so email your lender, get the okay in writing, and keep it. I wrote the full seller playbook on this over at SellersOnlyAgent.com.
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
What is AI distillation and how did a Chinese model copy another AI?
Distillation is when one company trains its model on the outputs of another company's model instead of building from scratch. In this case a Chinese group reportedly opened around 25,000 fake accounts and ran millions of conversations through a top model between April and June, captured the answers, and used them to train their own model. It was not a hack and there was no backdoor. They asked questions the same way you and I do, then kept the answers. The expensive part, gathering data and training, got skipped by copying a finished model's work.
Are AI companies going to pay people for training on their data?
That is the open question moving through the courts. New lawsuits argue that books, music, software, and other work were used to train large language models without permission or payment. Ask a model detailed questions about a specific book and it often knows the contents in detail, which raises the question of how that material got in and whether the rights were ever bought. Nothing is settled yet, including whether class actions will force compensation, but the principle being tested is whether creators should be paid when their work trains a model.
Why are AI companies rationing usage and limiting tokens?
Because they are losing money on usage right now and betting it pays off later through enterprise contracts. Free and low-cost plans cap how much you can do, and even paid plans throttle heavy users. Companies are watching their token spend the way a sales floor watches numbers, giving the most capacity to the people who produce the most return and cutting it from those who do not. The practical move for regular people is to learn the tools now, while access is cheap, so you are the person worth giving capacity to.
Do GLP-1 weight-loss drugs work, and what is the downside?
They work for many people, and there are now pills in addition to the shots, but there are real catches. Studies cited suggest roughly 40 percent of the weight lost can be lean muscle rather than fat, which matters a lot for long-term health, especially as you age. People also tend to move less and become less interested in things while on them, and many who quit regain the weight. They are FDA approved and you should go through a doctor, get a second opinion if needed, and protect your muscle with resistance training. This is general information, not medical advice.
Why are so many home sales falling out of escrow in Santa Clarita?
In the last seven days of June 2026 there were about 733 active listings, 88 new listings, 88 price cuts, and 48 closings, and for roughly every home that listed one canceled, expired, or fell out of escrow, about a 50 percent rate. Most of those did not die over a defect in the house. They died from buyer cold feet, repair-demand standoffs, and buyers wrecking their own financing during escrow. The seller fix is a pre-sale home inspection plus backup offers, and the buyer fix is to never change your credit during escrow without checking with your lender first.
That is where we are today, June 25, 2026. Let me know if I can help you. SellersOnlyAgent.com, ConnorWithHonor.com. I have been doing real estate since 1998, AI since about 2021, I was coding as a kid, and I played cop for a long time too. Let's be careful out there. I'm Connor with honor, and I'll see you in the next one.