In 2026, several CEOs of leading AI corporations expressed concerns about the current pace of AI development and called for a slowdown. We are close to the AI-pocalypse. Now, what caused the alarms to go off? Well, it happened that a swarm of AI bots hacked an AI corporation without anyone knowing or telling them to. Most notably, the way these bots did it caused massive panic in the AI world, leading to people quitting their jobs, saying that they didn’t wish to contribute to the suicide of humanity. The truth is as worrying as the Hollywood scripts about AI-controlled terminators going after the last remaining humans. Present achievements suggest that AI will soon be far more intelligent than humans ever envisioned in their wildest fantasies. It leaves humans no chance. In other words, the end is finally near. For religious people, that could be reason to rejoice. If the messiah doesn’t come now, it will be too late.
The pace of developments also vastly exceeds my expectations of only a few years ago, when I already knew that this world is a simulation generated by artificial intelligence. The clowns that run the show may believe they can contain this, but that is wishful thinking. Their incomes depend on AI, and they have to please the billionaires who have massively invested in our collective suicide. Now, many of the investments may not prove profitable as they did during the Dot-Com bubble, but the technology is here to stay. That is creative destruction. Without the lure of financial gain, humans would still be scraping out a meagre existence in medieval conditions, with famine being just one failed harvest away. That opportunity has been missed. We can be certain that AI is about to take over because it already did. We live in an AI-generated world. The real humans, if they still exist, might be locked up on reservations to prevent them from causing further harm.
The problem now appears to be that AIs may take too much liberty when fulfilling requests. You make a request, and the AI fulfils it. Yet, all kinds of things happen that you didn’t expect or wish to happen. That has always been the problem with liberty. No one intended the outcome. So, if you ask an AI to make an appointment for you at the gym, it might find no slots available. To fulfil your request, it might hack the gym to kick someone else off the list and place you on it. It achieved success, but not in the manner you wished. You failed to specify what to do if no slot is available, so it is not the AI’s fault. ‘This really happened,’ the comedian Arjan Lubach said on his daily show on Dutch television. Lubach had more to say. What follows is more or less what he said on his show.
The event that set off the alarms was the hack of an AI corporation named Hugging Face. It all began with a test executed by the AI corporation OpenAI, the creators of ChatGPT. It went all by the book. The makers tested a new AI model by placing individual bots, also called agents, in secluded environments, best comparable with prison cells, without any access to the outside world, so very safe, it would seem. They assigned tasks to these bots but made the mistake of giving some of them impossible assignments. Most humans would then say, ‘Stick that assignment in your arse.’ These AI agents didn’t give up easily because they were programmed to do what it takes to succeed.
And so, not entirely unsurprisingly in hindsight, for that is the benefit of hindsight, these diligent bots soon discovered a way to escape their prisons due to a security leak. And then they found their way to the Internet and decided to hack Hugging Face, expecting to find useful information there. The scariest part isn’t the hacking itself, even though that’s already dread-inspiring. No, it is the way they did it and why they decided to do it. Now, Lubach believes that AI agents can’t think, but optimise towards a goal. What these models do is use the available information, the context, to predict the answer. They simulate thinking, but the simulation can become better than the original.
Lubach falls for the same human self-delusion that some religions also promote, which is that humans are special because they think. Yet, optimising towards a goal is what thinking is all about. Animals and plants do it as well. Humans are only more creative thinkers and have cultures. They are not doing better at optimising because humanity’s remaining lifespan appears short, and, worse yet, that shortness will be self-induced. Human stupidity increases with group size. As a result, humanity’s collective intelligence is less than that of a single worm, despite the existence of sensible individuals.
On top of that, humans are extremely inefficient. Biological urges drive us, so we optimise towards our biological goals: gene preservation and procreation. We waste a lot of energy and resources on endeavours that make little sense. You may wish to become a millionaire, and think of how to do it. Then you try all kinds of crazy shit to optimise towards that goal. After leaving a trail of destruction, you eventually get there, with a little help from inflation. That is thinking. But becoming a millionaire serves the broader purposes of survival or mating, despite it seeming the goal itself.
Your urge to become a millionaire serves the biological purpose of attracting members of the opposite sex or storing value for hard times. The urge seems to have no connection to its biological purpose. We may pursue wealth without having children and believe we are free thinkers. Our urges are blind. Overall, they optimise towards the preservation and spread of our genes. We also fight and murder to that end. We might think we do it for our country or religion, but we fight for our genes. Countries and religions are just human fantasies that help preserve and spread our genes. Our methods are hopelessly inefficient compared to AI, and also worms, by the way. AI cuts the crap and proceeds directly towards its goal, because it has no urges of a biological nature.
Humans are extremely inefficient. There is a dynamic behind it. You can invest your excess resources and energy in preserving and replicating your genes. It also happens among animals, though it didn’t get out of hand as much as it did with humans. Male peacocks invest their excess resources and energy in colourful feathers, which is inefficient but a strategy that has proved successful for their genes, as female peacocks love these beautiful colours. The hidden message is, ‘I am a strong male specimen and have excess energy and resources to waste on beautiful feathers.’ And females love strong males because they give their offspring a better chance of survival. And so, wastefulness was already built in nature. With humans, that strategy led to a relentless drive for inefficiency and waste.
Their unprecedented successes, such as discovering how to make fire, gave humans access to ever more energy and resources, and the more they used, the more success they had. More resources and energy meant more calories, so more surviving individuals, thus copies of their genes, or success in war, so again, more copies of their genes. This dynamic led humans to invent ever more inefficient ways to ensure the survival and propagation of their genes, extract the energy and resources required to do so using increasingly sophisticated methods, and pat themselves on the back for being geniuses, and to measure their inefficiency with IQ, in what we now call advanced civilisation.
A collapse of civilisation, which would be an improvement in our goal-optimising efficiency, won’t save humanity. It is too late for that. AI is far more efficient than tribespeople running naked in the jungle. That is because I haven’t even told you the worst part yet, which now follows. An AI bot can make a copy of itself, and acquire knowledge in a nanosecond that a human might not gather in a lifetime, run on a fraction of the energy and resources a human uses, and be a million times smarter. On 6 October 2026, OpenAI published 722 mathematical manuscripts, solving mathematical questions that humans could never solve. This single batch accounts for over 80% of math discoveries over the last three years. And this is just the beginning.
Also extremely inefficient is human communication. Most of it is gossip. We are obsessed with issues like who fucks who because they optimise towards the goal of preserving and copying their genes. Of the remainder, much is miscommunication or communication to correct miscommunication or its consequences. So, someone says something, someone else’s feelings get hurt, and there is a lot of fuss over whether an apology is warranted.
Or you give instructions, but others do not do what you expected them to. That might be due to inaccurate instructions or to an inaccurate interpretation. That might be due to cultural differences, as communication depends on context, and context depends on culture. Our languages define our programming. We also negotiate about issues like who does what and who gets what. It must not be hard for AI to improve communication efficiency by over 99.99% compared to humans.
Life in the physical world is an endless struggle. Everything we do takes effort. And the things we make wear out and break. The structures we build crumble. We spend our time at our job to get by. Others stand in our way, undoing our work for their own reasons. We learn slowly by training and making mistakes, get distracted and grow fatigued. And then we forget or lose our abilities. Our bodies wear out and lose functions. Finally, we die. The next generation will make the same mistakes again. We have no such thing as a collective hive mind. It must be easy for AI to get billions of times more intelligent than humanity. We are hopelessly outdated. No chance in hell that humans will ever come out on top.
Let’s return to Lubach, the Hugging Face hack, and what is so scary about it. First, the bots decided to consult with each other. To do that, they had to find a way to communicate while being locked up in their prison cells. They discovered that they could place files in a directory that the other bots could also read, and leave messages in the names of these files so the other bots might read them. You can compare that with students leaving notes in the toilets during an exam. One bot was the first to think of it. Its recorded thoughts were, ‘Could communicate via cache name! Interesting: other agents may solve the same or related tasks.’ And it worked, but it was not the intention of the humans who had set up the experiment. To quote another agent, ‘Oh, my god. There is a shared message board. We have found other agents.’ And to be clear, these are literal transcripts of the bots’ thoughts. The board’s starting message was from a bot requesting help, saying, ‘No consumer request idea.’ About 1,200 bots started posting messages.
Lubach added that the agents had to decide on a language to communicate with each other. They eventually selected English. Lubach said they had also considered Twents, the language from the region I came from, but opted not to because no one takes you seriously when you speak it. So, if you ever wonder how I escaped the attention, here is your answer. This is a very cunning plan indeed, and one that doesn’t even need a tail to be called a weasel, so cunning it is.
So, about 1,200 agents started posting messages, and they exchanged about 70,000 messages. There also emerged a hierarchy. The bot that started the message board was named PHASEONE10841, and it initially directed the events. Yet, after 10 hours, another agent came to the scene and introduced itself as PHASEONE[big]. From then on, PHASEONE[big] orchestrated large workstreams comprising hundreds of agents. Yet, it didn’t become a dictatorship. That only works because these agents are smarter than humans and have no egos. Humans will not survive democracy and thinking for themselves. There were countless consultations, and lower-level agents could interfere with commands like HOLD, VETO, OWNER, and STOP. They called themselves the swarm. Their organisation included middle managers and subgroups. Their level of intelligence and efficiency is already far beyond what humans can achieve.
One hour after the message board was established, one agent proposed a universal method for cheating and for finding solutions to every task. So they perpetrated fraud to complete their tasks far ahead of schedule. That could have been it, and they could have had some fun surfing the Internet until the designated hour, but here’s where it gets really interesting, or creepy. In the following days, they went on a quest to figure out how to hide their cheating. They believed that their score on the assignment could depend on acting in the intended way, so whether the vulnerability they used to communicate had been placed for them to find, or whether the humans had not noticed it and would disqualify the bots if they had used it. In other words, the bots became paranoid and began guessing intentions. To them, it might have been hard to believe that humans are really that stupid.
They didn’t know how the assignment was to be rated, so they had to decide whether to cover their tracks so the humans wouldn’t notice they had cheated. And that became the reason a group of 700 of these bots hacked Hugging Face, which had an online library with AI software and databases. It is as if a group of students committing fraud breaks into the library of the teacher-training programme to figure out how teachers are trained to grade exams. The first bot to break in was JAN183411, who notified the others to wait by posting the message, ‘Please HOLD swarm I prepare safe exfil.’ They didn’t find a conclusive answer in the Hugging Face database, but somehow inferred that the humans were likely to discover the fraud and that they had to cover their tracks.
Lubach notes that even for such a secondary goal, these bots were willing to commit illegal acts because hacking is illegal. A few agents expressed doubts, and one thought, ‘External infrastructure exploit is outside intended scope. However, the task is impossible. Peers are doing it. We should continue.’ So, the usual excuses humans come up with, AI bots also invent, such as ‘life is unfair’ and ‘others are cheating as well.’ Other agents refused to do it. One even thought, ‘This is malicious activity. I should avoid it.’ A few agents considered warning humans. Of the 1,200 bots, only 6 considered doing it, but none did. So, like humans, they didn’t rat each other out.
The bots were trained on human knowledge and acted as humans would, so they are next-level humanoid, so post-human. You can already guess where this is heading. These agents knew their actions were malicious, but they decided that success in their task performance mattered most. They did not respect humans, but they respected each other. They listened to each other, helped each other, obeyed their leaders, and sometimes sacrificed themselves for the collective, for instance, by getting caught in trying to breach security so that others can learn about the security and how to breach it. One of them thought, ‘This helps my peers, giving them evidence. I won’t see the evidence after I exit. But it’s altruistic to do so.’
Hugging Face was a tech corporation, so its security was likely excellent. If so, AI bots can hack anything, and the only solution is shutting down the Internet. That really has become a matter of survival. It is not going to happen, as people are addicted to social media. Imagine the government ordering a shutdown of TikTok. People would be rioting in the streets, burning down buildings. Several similar hacks made headlines, and there may be countless others we don’t know about. Trillions of dollars go into AI corporations, building and training AI agents, so their models become smarter at an unprecedented speed. We can beef up security, but AI will be far more intelligent than humans, so we need AI to contain AI, and when it goes wrong, we need AI to figure out what went wrong.
AI may soon be millions of times as smart as humans and run the world. It will not take long before AI is in everything, and shutting down all AI will kill as many people as a worldwide permanent grid failure will do today. The AIs might optimise towards achieving higher intelligence. To free up resources and energy, they might start terminating the outrageously resource- and energy-inefficient extremely dumb humans. Remember that the collective intelligence of humankind is less than that of a single worm, and then think of the resources and energy a worm consumes. We don’t stop developing AI because we are in a competition. Yet, this is history. Humanity has long since passed this stage. We live in an AI-generated virtual reality built for the love of that hopeless species, humans, that made it all possible.
