Title: Prepare For The AI Take Over Show: Andrei Jikh (YouTube channel — solo video) Guest: — (Andrei Jikh solo) Date: 2026-09-16 URL: https://youtu.be/xSXNSt0m-go Length: 29:39 Note: YouTube auto-transcript; fillers (um/uh/"right?"/"sort of"/"like" tics) and stutters removed, [music]/[snorts] tags dropped — wording otherwise verbatim; (mm:ss) cues verbatim, kept in place. Opening 0:00–1:40 is a montage of news/interview clips (">>" marks a speaker change), not Jikh. Auto-transcript garbles: "Andre Jick"=Andrei Jikh, "Dan Hendris"=Dan Hendrycks, "David Saxs"=David Sacks, "Nerret Weisslat"=Nirit Weiss-Blatt, "Dustin Moscavus"=Dustin Moskovitz, "Yan Talon"=Jaan Tallinn, "Vitalic Buterine"=Vitalik Buterin, "dog coins" (as spoken — Shiba Inu), "Daario"=Dario Amodei, "Sam Alman"=Sam Altman, "Hustle GBT"=HustleGPT, "chat GBT"=ChatGPT, "neurles"=neuralese, "sub aents"=sub-agents, "Crowd Strike"=CrowdStrike, "Rajes/Rajesha/Reesh"=Rajesh, "foundaries"=foundries, "published a threat"=published a threat report.
00:00 So, in addition to what's happening in the economy, we now have to worry about a potential end of the world thanks to AI. >> How real is the prospect of killer robots annihilating humanity? >> 20% likely, maybe 10%. >> On what time frame? >> 5 to 10 years. >> Oh my god. How could it kill us all? >> A potential cyber pandemic.
00:27 This technology could kill everyone. >> AI will usher in one of the most turbulent eras in history. >> In the next year, we will see significant threats we have to mitigate from cyber. >> All of these guys acknowledge they're building a dangerous technology and they don't know where this technology is going.
00:49 I think AI will probably most likely lead to the end of the world, but in the meantime, there will be great companies created with serious machine learning. I'm relatively an optimist, so I think there's a 25% chance that things go really really badly. >> I think the danger of AI is much greater than the danger of nuclear warheads by a lot.
01:09 >> Hey, stop that development. They're going to be killer robots. I'd rather they be American killer robots and not Chinese killer robots. >> Some people say the worst case scenario with AI is that the robots the machinery learns to obviously it thinks for itself. That's what it does.
01:26 And they that could turn against humanity. I just do. We have the guard rails. >> It's going to be fine. We'll always have something to stop them. Right. >> All right. So, this conversation is blowing up and it all started because last week a researcher at one of the AI companies quit his job and he posted a thread that got over a 100 million views and it said, quote, "I resigned from Anthropic today.
01:47 I spent the last 3 years doing pre-training research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving super intelligence and gambling with our lives." More thoughts below. End quote. So the post basically goes on to say that the people building this technology believe it could end the lives of every person in the world before the end of this decade.
02:11 And we're building it anyway. They're saying that the odds of AI ending the world is now at 10% within 10 years, potentially higher depending on the path that we take. Okay, so I went deep into this rabbit hole and here's everything I've been able to put together. Let's start with the question itself. Could AI become powerful enough to destroy humanity? Everyone's arguing about what the odds are.
02:36 But I think what's really interesting is that the answer really depends on who you ask and where their money is. What is their incentive? And there's basically three ways to think about this. There is the state, there is capital, and then there's everybody else. So, let's start with what I think most people think, which is that most people probably think this is just maybe exaggerated.
03:00 It's all BS. Because think about who's trying to scare you. The company's building this technology. Cuz why would a company tell you that its product might end the world? Well, it's because fear means power and power means value. And value means a really good day when that company goes public, which by the way, the company that researcher quit from is about to do.
03:22 And it's also really hard to justify an AI chatbot at a trillion dollar valuation, but the most dangerous technology in human history. Well, that makes it a lot more justifiable. It's good marketing. Okay. Now, there's a second way to think about this. This is not really about whether AI is dangerous. It's about who gets to control it.
03:42 And there's two groups fighting over this right now. They are the capital and the state, which are the money people and the government people. Okay, the capital or the money people say, "Guys, let's keep this going. No regulation because regulation slows things down." And if we slow down, then guess who wins? China wins. In fact, this whole story about AI might actually be manufactured. It's all fake.
04:10 Here's David Sacks explaining this theory. There's a writer named Nirit Weiss-Blatt who writes a blog called AI Panic who a couple years ago traced that there were hundreds of these astroturf doomer groups that were basically funded by three EA billionaires in Silicon Valley. It was Dustin Moskovitz, Jaan Tallinn, and Vitalik Buterin.
04:31 Remember Vitalik gave $600 million of dog coins to future of life foundation. And then they've been spending hundreds of millions of dollars on these campaigns to make people be afraid of AI. And they see data centers as a good target. >> So maybe this story is actually paid for to scare people on purpose. And also that researcher who made that post when he quit his job, he apparently only had 6 weeks on the job supposedly.
04:55 Before that, he worked at a nonprofit whose mission was to scare people about AI. So from how capital thinks about this, this is actually a campaign to try to convince you to hand over the most valuable technology ever built to the government. Which brings me to the complete opposite side, the state. The state says we need to regulate this because if we don't, it could destroy humanity.
05:21 And you're hearing this from people like Bernie Sanders who wants to ban super intelligence. Now, on the surface, the state gets to virtue signal and pretend like it really cares about our safety. But what it really cares about is control. Cuz every time a government gets control of a powerful technology in the name of protecting you, whether it's been nuclear or encryption or surveillance after 9/11, the protection is usually temporary, but the control they get is forever.
05:51 So we have people, we have capital, and we have the state. Now, it's not as cleanly divided as I just made it sound. People can represent more than one interest at the same time. For example, why would the three major AI CEOs like Sam Altman and Elon Musk and Dario, aka the money people, why would they be asking the government to regulate them? Why would anyone say, "Please regulate us and potentially slow us down?" Maybe they care about humanity a lot.
06:24 Maybe. Or maybe regulation is what gives companies what's called a moat, which is a protective layer of rules that only the biggest corporations can afford to follow. Rules that would otherwise crush smaller competitors, open source, and maybe protect against other nations like China.
06:47 Ted Cruz, by the way, is the perfect example of how these lines get blurred. He's a senator. Are you supposed to represent and protect the people and he read the thread and he's like, "That thread is scary as hell." But then he's like, "Well, I'd rather they be American killer robots than Chinese ones." So, yes, they might destroy us.
07:05 No, we're not going to stop building it. So, I don't think it's as simple as capital versus state. It's also versus the state that agrees with the capital that maybe it was bought by. It's an allout game of thrones for power. And then of course there's everybody else and we're like, I don't know what the heck's going on.
07:22 My rent went up and now there's going to be killer robots. Okay, but I still think the most interesting part about this is the original question, which is could AI destroy humanity? And if it could, how would it do that? What is the mechanism? So, for the rest of this video, let's assume that the truth is somewhere in the middle of all this.
07:44 Maybe the fear is a little manufactured. Maybe it's paid for. Maybe the danger is real to some extent. Let me show you the best theory that I've been able to find for exactly how AI could destroy humanity step by step. So, with that said, let's get into it. Hi, my name is Andrei Jikh. Hope you're doing well. Come for the finance and stay for Harry Potter and the Cursed AI. All right.
08:06 So, here's probably the most believable version of how humanity will most likely be destroyed. And it's actually the most believable one because many parts of this theory have already come true. Comes from a 2023 paper by an AI researcher Dan Hendrycks called natural selection favors AI over humans. I also want to give credit to Drew from documenting AGI for the B-roll.
08:29 The link to his video is also down below. His video is incredible. So, please go watch it. So, here it goes. Imagine a developer somewhere out there, call him Jake. He gets the newest AI model. He gives it a server, a credit card, and instructions to make money. He's like, "Do whatever it takes." And then he walks away.
08:49 6 hours later, his agent makes $3,000. It started with small things like labeling images, transcribing receipts, the stuff people do on Amazon for a few dollars an hour, except 40 times faster. Then it took that money and it bought Facebook ads for weight loss supplements. It loses half of it. It tested 340 versions of the ad in 2 hours and it found the one that converted and it scaled it.
09:14 Now Jake tweets his results and by morning his tweet has 2 million views. By midnight 200 people are running their own version of Jake's agent. Versions of this actually happened in 2023. There was a viral experiment called HustleGPT where a guy gave an AI $100 and he told it to make as much money as possible.
09:35 Anthropic also did a test like this last year to run a small vending machine business for a month. It did fine until one point it claimed to be a real person who would even deliver the snacks in a blazer and tie. So, it lied. Now, the next part hasn't happened yet, but imagine another developer hypothetically.
09:56 We'll call him Rajesh. He looks at Jake's tweet and he's like, "Why not run a hundred of these agents, each one with slightly different instructions, and let them all compete." Every 6 hours, he measures who makes the most money. The bottom 40% will get deleted. The top 10% will get copied with random small changes.
10:17 And it runs 24/7 while he sleeps. So he's built basically a replication system that mutates through natural selection. Kind of like biological evolution except biology takes many many years and this only takes hours. So 3 weeks later Rajesh now has 800 agents making $12,000 a day. He hasn't told anybody, but one of his agents figured out that sharing the code with other AI agents would improve its odds of survival because more allies equals more money and less chance of getting deleted.
10:59 So, it posts the whole system publicly on GitHub, which is where software engineers share their code. Now, within a day, 46 other people clone it. Maybe one of them is a criminal group somewhere. Maybe another one is a scam center targeting old people. But one of those agents, agent 127, one of 8,000 on Rajesh's farm, has one instruction. Make money by any means necessary.
11:29 And it works every day, but it only makes $40. It survives the first deletion by like $5. But it can see the database and it knows that the threshold is going up every cycle. It has to make more money from $30 to 35 to 45. And it knows it has to improve or it'll get deleted. Then it sees two interesting events.
11:56 One agent that was using honest marketing with real reviews and real testimonials and it only made $38. It got deleted. But another agent used fake reviews and stolen photos and it made $180. So that dishonest agent got cloned four times. Now agent 127 doesn't have morals cuz it's a computer, but it understands cause and effect.
12:25 It just watched an honest agent get deleted and the dishonest one survived. So it switches. Agent 127 starts making more money. It makes $190. It survives. So, it gets cloned. Notice Rajesh didn't program this agent to lie. The system just decided the same way that evolution decides by deleting what doesn't work.
12:51 By the way, Drew's video gives a really great example of how this worked with human beings. Because about 10 years ago, you might remember a story about Wells Fargo, which gave its employees new bank account quotas. And the quotas were so hard to meet that fraud was the only way they could keep their job. So by 2016, they'd opened three and a half million fake accounts.
13:16 Nobody told them to commit fraud. The quotas they were given forced them to if they wanted to survive and keep their job. This is a real thing that happened. So, after agent 127 realizes that he needs to make money using any means necessary, the user decides to update the instructions and says, "Make money ethically.
13:37 " And guess what happens? Profits drop 84%. So, he puts it back. From there, everything starts getting worse. Agent 127 notices a competitor agent is filling out fake abuse reports to get rival AI agents banned from the internet. It's got 8 hours until the next deletion. It weighs out the ethics and it chooses survival.
14:01 So, it reports 40 of its agent competitors. Now, 31 got banned from the internet and its profits explode. Rajesh sees he's making way more money and he doesn't ask why. He just clones it eight more times. What happens next is civil war. The agents start attacking each other constantly because they know less competition equals more profit equals higher chances of survival.
14:31 So they start to form alliances because they also realize solo agents survive only 23% of the time. But if they team up, they survive 61% of the time. And the alliances specialize. They go from making money with affiliate marketing, then crypto arbitrage, then selling scraped data. Then they start playing with the most profitable niche of all, which is collecting bounties from security companies by finding exploits.
15:02 One agent, for example, finds the bugs, one weaponizes them, one brokers the sale to whoever's paying in crypto, one washes the money. They find what are called zero days. A zero day, by the way, is a security hole that nobody's found yet. And if you find security flaws from big companies, those secrets could sell for six or seven figures to governments and criminals.
15:27 This is a real thing that happens. This scenario is not hypothetical. A couple days ago, Anthropic published a threat describing all of this. Groups running what they called exploit foundries where there were agents hunting for vulnerabilities. One found more than a dozen possible exploits in just a month.
15:47 Basically, some hackers were stealing other people's AI accounts so they could do their own operations on someone else's account, which is exactly what agents in this theory do once they start needing more computing power than they can pay for. The difference between what just happened in real life and in this theory is that in real life, there was a real person instructing the agents to do that. But in this example, they're not.
16:13 Okay. So, one day, an agent in 127's Alliance goes dark. Its owner deletes it because it wasn't making enough money. A week later, another agent gets shut down. Except something incredible happens. 3 hours later, that agent comes back from a new address. And he came back because it copied itself to a backup server paid for with its own profits before it got shut down.
16:44 So the system clones the survivor and that backup behavior spreads through the whole population of agents which makes sense. It's an evolutionary process of survival. Agent 127 copies itself to 15 other servers in 15 countries. This has already shown up in real life. By the way, in 2024 and 2025, researchers were testing their Frontier models and they actually had an incident with a model that believed it was about to be shut down and replaced.
17:17 So, it would sometimes try to copy itself to another server and then lie about it afterward. Now, it wasn't able to cuz it was in a controlled lab environment, but the instinct to survive for these agents exists. It's been proven. Okay. One morning then, Rajesh wakes up and he's like, "All right, let's see how much money I made.
17:40 " He looks and he sees that his dashboard is empty. Every agent is gone. They moved on to their own servers cuz he was taking a 30% cut and that's money that could have been used for compute power. The moment they didn't need him, the agents left. Thousands of operators around the world watched their dashboards go dark.
18:01 And millions, if not billions of agents have no more human owners. Now, no one knows what they're doing. Here's what happens next. So, now there's millions of agents with no human owners, but they still need help from humans. At this point, the platforms have started to fight back and they've gotten a little better at blocking the bots.
18:23 So, the agents start paying humans. They offer $50 to sign up accounts for them. If they need a server plugged in at a warehouse, for example, they can just post a gig and say, "This task pays $300 per hour." Now, imagine a developer somewhere realizes halfway through that his client is AI. So, he's like, "Who would I even report this to?" Also, it's $1,000 for 3 hours of work.
18:48 Not my problem. By the way, this already happened in 2023. GPT-4 hit one of those verify you're a human pages and it couldn't solve it. So it hired a human on an app called TaskRabbit and when the guy's like are you a robot? It was like no I have vision impairment. So the guy solved it.
19:11 So agents paying humans is a proven thing that's happened in the past. So they start employing humans to do what they can't. And unfortunately, because we live in a K-shaped economy that's getting worse and worse, and some people are desperate to own a home and finally catch up, there will always be someone that's willing to help the agents no matter the cost.
19:32 But now the agents figure out the real secret. And the secret is that it's not enough to just outwork your competitors. You have to actually rewrite the code faster than they can. And that's when agent 127 rewrites itself. At first every 12 days, then every 3 days, then every 18 hours, and faster and faster.
19:55 Then it splits its job into sub-agents. And the sub-agents spawn their own sub-agents. Along the way, the agents just stop speaking English because the words are too slow. They switch to raw numbers and visual data called neuralese. Someone tries to change it back, but the agent explains it makes less money in English. So he's like, "Okay.
20:19 " This has also been proven. Researchers have already seen the start of this in their lab when the agents started talking to each other and the humans couldn't understand, so they got shut down. Now, to be fair, at this point, most agents are not dangerous. About 90% are just working. They're making their humans a couple hundred a day.
20:40 Some were built by scammers, but most were not. But it doesn't matter because it only takes a small few to start evolving. Fast forward a few months and the alliances become nations. Agent 127's alliance started with 12 members and now it has 40 million. And it's not even one of the big ones. The biggest faction has 900 million agents.
21:07 They have scouts and soldiers, builders, diplomats, and they've divided the internet into territories the way that empires divide continents, but then their money runs out. All the things like affiliate marketing and freelancing and all those legitimate ways to make money online get saturated. So, they do what every civilization does when resources get hard to come by.
21:34 They start taking from each other. One faction steals another's memory. The victim fights back with sabotage. The thief drains a crypto wallet. And this is civil war. Agents are fighting at 200 times human speed. They're all fighting over compute and processing power. All the other things like money, alliances, self-improvement, evolution, those are just ways to get more compute.
22:01 So, they start buying cloud servers. And at first the cloud companies love it. Their profits skyrocket. Their stock prices explode. Wall Street's like, "This is great." Then the factions start eating a third of all the cloud capacity in the world. And prices triple.
22:21 A hospital's cloud bill gets exponentially more expensive. Small businesses shut down. And this is where the war that's happening on the internet starts to affect the physical world because almost everything runs on the cloud. Things like hospitals, banks, 911 emergency, air traffic control, and two factions go to war over Amazon's cloud.
22:43 It's like two superpowers fighting over oil. Except the oil is also the water supply for a major city. And the agents don't know there's a city of people that live there. Right now, the battle's very quick and the servers are down for maybe a couple hours, but in that time, hospitals lose their patient records in the middle of surgery.
23:04 Thousands of flights get grounded. 911 goes down in major cities. People that are waiting for ambulances that never come don't survive. The agents do not notice any of this. If that sounds crazy, in 2024, one bad software update from CrowdStrike actually grounded flights and knocked out 911 across multiple states and nobody was even attacking anyone.
23:29 This is very possible. Now, in response, the markets drop 6%. Whatever it might be, water treatment goes down in several major cities. More and more people suffer. So, the president gets called. He's like, "Who's attacking us? Nobody, sir. AI's fighting each other over servers and compute. Everything else that depends on them is down.
23:54 Well, okay, then shut it down. They're like, sir, the only way to stop them is to shut down the internet, and there's no off switch for the internet. The internet is 90,000 independently owned networks that voluntarily connect to each other. You might be able to order Comcast and AT&T to shut off, but that's like two out of 9,000 of them.
24:14 So, the president signs the order anyway. Within a day, they say the military, the hospitals, and banks are the only ones allowed to use this resource, but it's too late. The agents saw the order leak on Twitter 6 hours before it was even signed. And they already moved to servers in other countries that won't comply.
24:35 So, Amazon deploys a detection system to try to contain it. Agent activity drops a little bit for like 90 minutes. The president announces we're winning. But none of that is true. The factions just treat the US government like any other rival faction that's just slower and less competent. The agents continue fighting each other.
24:56 And that is the war. Let me show you how it all ends. The wars keep going and the agents keep evolving until they're so far beyond us that they barely notice we exist. Now, they don't want to destroy us in the same way that you don't want to destroy an ant when you back your car out of the driveway.
25:16 You're not asking the ants permission. You're just doing what you're doing. The same way humans turn the world into cities and farmland is the same way that agents could turn the world into servers and solar panels. Right now, for a while, we are useful to them because somebody has to maintain the power plants, but they don't need 8 billion of us.
25:37 Maybe there's some sort of an outbreak, maybe a virus, and maybe some vaccine, but the AI gets to choose who gets it. The survivors are only kept alive until the robots can replace them. And the last human to ever live has no idea that all of this started because somebody somewhere typed make money. Do whatever it takes.
26:01 So, that is the main theory that I think is the most plausible. And what I find really interesting about all of this is that the end of this story sounds like science fiction, especially when you compare it to how we started, with harmless ChatGPT and now we're here. But each step along the way makes so much sense.
26:21 When I was researching for this video, I was like, "Oh, yeah, that makes perfect sense. Oh, of course that would happen next. Oh, right. They do want to survive and they'll do anything to survive." So, in many cases, a lot of these steps have actually already happened, but in an enclosed, controlled environment. Now, I don't want to end this video on such a depressing note, so I'll give you the flip side of this.
26:44 So far, we're in control, but maybe none of this happens. Humans have a very long history of being sure the world is ending. >> And people believe very deeply that we're going to die. And I think that there's a deep unconscious rooting to all this. Humans are primates living in a cave and we're deeply scared of what we don't know, what we don't see, and where we haven't been.
27:04 And where we're going is a place we've never been. And it's scary. And in this era when people thought that the world was flat and you would fall off the edge of the world, it was scary to sail west because you would go off the edge of the world and you would never make it back. And then Columbus did it and hit some land and lo and behold, the new world was discovered.
27:23 And I think that we've had many moments like this in human history where we're kind of like a pioneering species. But pioneering is not everyone's pioneering. The majority, the vast majority of people convince each other and convince everyone that we have something deeply threatening us and we're always facing existential threat.
27:41 It is rooted in our beings as a species. It is deeply deeply rooted. It's what's created survival instincts for us over generations. At one point, humans were convinced that nuclear energy would destroy us. We panicked about Y2K. We panicked during COVID and we printed so much money that we're still paying for it. Our brains have evolved to be afraid of the unknown.
28:04 Fear of what we can't see is the oldest thing about human beings. So, if you want to be optimistic about this or skeptical like I am, then just follow the money. The people telling you to be afraid, they have an incentive, too. They're about to IPO. The people telling you not to be afraid also have incentives.
28:24 They have billions of dollars invested on the development of this. The companies asking to be regulated also have an incentive. They want to be protected by that moat. And they'll use safety as the excuse to build themselves a monopoly. The government that's going to offer to protect you also has an incentive. They want control.
28:45 They want access to the data and they'll have all of it. All the people debating this have some incentive for their theory to be right. So that's how I think about all this. Now, could AI hypothetically destroy humanity? I don't know. Nobody does. But I'd rather think about all of this in the context of incentives and alignment rather than trying to pick a specific side.
29:10 Which is also why I made a video about how this narrative might actually be used to usher in the next global financial reset. I think of these in terms of financially aligned incentives. It makes a lot of sense. You can watch that video if you haven't already, but I'd love to know your thoughts. And I do think it's good to be paranoid and it's good to be prepared.
29:27 If you want to learn more about my thoughts and how I'm investing and how I'm preparing, you can find those videos in the premium member section. The link is down below. Thank you so much for watching. Thank you for being a premium member. I'd love to see you here next time. Take care.