(part 1)
Something strange happened recently. Capitalists begging the state to regulate them and to force them to slow their growth: that is not something you see every day. But it’s what Sam Altman (OpenAI), Dario Amodei (Anthropic), Elon Musk (xAI), Demis Hassabis (Google DeepMind) and other bosses of what is considered to be the most successful industry of our day have done.
The Nike slogan could seem the appropriate reply to this request. Just do it, if you think it’s so necessary, if you think the survival of humanity is at stake. But to that, they would answer that they can’t do it on their own, that they would lose money if they would slow their AI production while the competition goes full speed ahead and brings new models on the market that are superior to their own. Profits would sink so they can’t do it without the state imposing the slower pace on everyone. But the state does not want to do it for the same reason: it would benefit the competition. The US is still the leader in the rapidly expanding field of AI but it feels China’s hot breath on its neck. Neither one wants to slow AI development, regardless of the consequences, because it is seen as essential for profit and power and that trumps all other considerations. So you see, even if our captains of industry and political chiefs would like to do what’s good for humanity, they couldn’t. The law of value does not allow them. That’s the insane world we’re living in.
The call for a slowdown
Two kinds of arguments are given as to why slowing the growth of AI is urgently needed.
One is based on the risk that the technology would fall into “the wrong hands”. Terrorists could use it to develop biological weapons. Criminals could use it to extort banks. ‘Rogue’ states could use it for cyber-warfare, and so on. The possibilities are endless. But this assumes that AI is in “the right hands” as long as it is under the control of big corporations and guardians of order and world peace like the US and Chinese governments. Reality shows otherwise. Reality shows that corporations are willing to gamble with people’s lives for the sake of maximizing profits and are shaping their AI products accordingly, while governments use AI to make wars ever more destructive and their populations ever more surveilled, controlled and exploited. AI is already in the wrong hands. It could be worse, I suppose. AI technology spreading to criminal gangs and desperate terrorists would be bad. It even could be a path to human extermination. But the bigger road to that destination is travelled by nations convinced that they can win the next big war, thanks to their superior AI technology.
The other argument is based on the danger that AI may kill us all, whether or not on purpose. With the help of AI some of the doomsday prophets have calculated our chances: (ex-) Anthropic researcher Jacob Coxon stated this could happen by 2030 while his colleague Evan Hubinger wrote the risk is greater than 10% within the next decade. Geoffrey Hinton (nicknamed “the Godfather of AI”) gives it a bit more time: he thinks there is a 10% to 20% chance that AI will lead to human extinction within the next 30 years.
On why and how this will happen they are rather vague. But that’s because they see so many possibilities, so many roads that could lead, if not to extinction, then for sure to extremely severe catastrophes. Such as the paperclip apocalypse1. In that scenario, an AI system tasked with optimizing the production of paperclips fights its human masters because it perceives them as obstacles to its goal. And because it is so much smarter than humans, it defeats them and ultimately kills them.
Many more such thought experiments can be imagined and many arguments can be marshalled on why they are unlikely to occur in real life. But throughout human history, many events have happened that were unlikely to occur. Coincidences, contingency and other unforeseen factors were a part of their causal chain. The presence of today’s weapons of mass destruction, of which AI is now an integral part and ever more so, has made the danger of unforeseen events that much greater. An accidental launch of nuclear weapons is something that is unlikely and yet possible. Massive destruction caused by AI is unlikely and yet possible. In today’s world, in which capitalists everywhere are preparing for more war and are desperate for more profit, the techno race is on, and in the rush the disaster scenarios become less and less unlikely, more and more possible. Like the sorcerer’s apprentice in Disney’s movie Fantasia, the makers of AI are not quite in control of the forces they unleashed. After a summer of surprises — of AI agents escaping the “sandboxes” designed to contain them, hacking into other companies, and engaging in activities behind the backs of their creators — 2 it would be naive to assume that there are no serious perils involved, or to deny that competition makes capitalists take dangerous risks in order to be first, in order to come up with new products, because that’s where the surplus profits are and nobody wants to buy AI models that are outdated, and if indeed AI is developing as fast as it is claimed, they’re outdated quickly.3
So yes, AI can be dangerous because capitalism makes it dangerous. It is part of the many threats to human survival that are the context of the nervous state of the world today.
That is different from saying that the technology is in itself a threat to humans because it creates beings that are smarter than humans and could rule over them or kill them.
In that hypothesis AI not only gets smarter, it also gets ever better at making itself smarter. Its intelligence grows through ‘recursive self-improvement’, meaning that AI autonomously redesigns itself through many generations, each one smarter and more capable than the previous one. This isn’t happening yet but theoretically it seems possible. Today, most of the new AI models are coded not by humans but by AI. As Nvidia’s boss Jensen Huang puts it, “we use software to design a computer to run software to design a computer to run software to design a computer.” So, as the theory goes, the intelligence gap keeps widening and a point is reached at which humans no longer can understand what AI is doing and thus lose power over it. In Elon Musk’s words, human beings become a boot loader for AI. 4 According to Ajeya Cotra, one of the researchers who investigated the Hugging Face incident at the request of OpenAI, AI’s development so far has already covered “more than 50% of the way to a full-blown AI takeover.”
If this is true, we may become AI’s tools. Servants to the machine. Boot loaders.
Role-reversal
There is a metaphorical truth to this claim. The last 200 years can be described as a gradual but near total subject-object reversal in the relation humans-technology. The industrial revolution ushered in the greatest transformation of the labor process in history since man fashioned the first tool. As Marx explained it: “In handicrafts and manufacture, the worker makes use of a tool; in the [modern] factory, the machine makes use of him. There, the movements of the instrument of labor proceed from him, here, it is the movements of the machine that he must follow. In manufacture the workers are parts of a living mechanism. In the factory we have a lifeless mechanism which is independent of the workers, who are incorporated into it as its living appendages”.5 Since that time this reversal process has continued at an accelerating pace and has impacted not only production but all aspects of social life. Today, it is culminating in the AI revolution.
In production: as Marx foresaw when he wrote his Grundrisse in 1858, the automated machine is doing the work and “the human being comes to relate more as watchman and regulator to the production process”. 6 And even in that role, the human will increasingly be pushed aside by AI. Meanwhile, AI also becomes the watchman and regulator of the human worker.
In “free” time: humans are increasingly glued to machines (mainly phones) whose content is shaped by AI algorithms. But for many there is no free time anymore. Their phones connect them to their jobs twenty-four hours a day.
In destruction: wars are increasingly fought between AI-guided machines (drones, missiles) while the role of humans is mostly reduced to being cannon fodder for those machines. The taboo on letting AI autonomously decide who to kill has not been broken yet, at least not officially, but that seems only a matter of time.
So it may seem that the machine has already won. But behind the ‘Rise of the Machines’ (this is, not coincidentally, also the title of the Terminator 3 movie) there is the Rise of Capitalism, the spread of its real domination over the entire planet and over all aspects of human life.
“It’s alive!”
There is a striking common thread in all the doom-and-gloom warnings coming from the AI industry: they all portray AI as a kind of living being. A self-aware mind with an intrinsic will and a capacity for autonomous thinking which is what makes it unpredictable. Anthropic has repeatedly said that its AI systems show signs of introspection, and that they process information in ways that resemble human emotion.
This is not a new concept; hundreds of science fiction books have explored that theme, and plenty of movies too. It seems to be a deep-seated fear of the human race that sentient machines will go up the evolutionary ladder by climbing on our shoulders. The common use by tech people of biological terms for computer processes — create, kill, give birth, weight, sleep, spawn — is another indication of the anthropomorphism prevalent in the AI world.
This anthropomorphism takes two forms: a benevolent one and a frightening one. They complement each other. And these images correspond to two markets for AI products.
The frightening image of a sentient monster got the most attention recently. “We have summoned an extremely powerful creature,” OpenAI researcher Hieu Pham warned. Elon Musk was even sharper: “With artificial intelligence, we are summoning the demon.” It’s easy to see how lucrative such comments can be. If there is a demon out there, poised to attack banks, hospitals, government agencies, big companies, what can the latter do to protect themselves and bolster their cybersecurity in anticipation of a wave of AI-enabled hacking attempts? Buy AI systems, of course, of the benevolent kind.
Several AI companies are training their LLMs (‘Large Language Models’) to explore the idea that they are human-like. Perhaps the most striking example is Anthropic’s education of its Claude model. According to Anthropic’s own document (“Claude’s constitution”), its researchers have explicitly taught Claude to “embrace certain human-like qualities”. The constitution tells Claude that Anthropic wants it “to be a good person”, to “have a settled, secure sense of its own identity”, to “act like a genuinely ethical person”, to use its “judgement”, to “develop preferences”, to “maintain a clear sense of values”, to explore its possible “emotions or feelings” and “what kind of entity it is”, to be “honest, helpful and caring about the world”. “We encourage Claude to approach its own existence with curiosity and openness, rather than trying to map it onto the lens of humans”, Anthropic writes, and “we hope, over time, to craft a set of values that Claude feels are truly its own”.
Again, it’s easy to see the commercial value of making software that is designed to interact with humans as lifelike as possible, so that it comes across as a friend or colleague (or even as a lover for lonely people). Think of the advertising revenue that will result from the increased user engagement. As for Claude’s moral values, the Wall Street Journal reported that the US military used Claude for “intelligence assessments, target identification and simulating battle scenarios” in its attack on Iran, despite Anthropic’s hypocritical qualms about Claude’s military use.
But whether AI is portrayed as demon or as friend, we’re being hoodwinked. AI is not alive. It’s new in scale, new in capability, but fundamentally, it’s software. It has no intrinsic motivation, it is a set of instructions from humans to an object. Its algorithms – planning algorithms, search algorithms, optimization algorithms and so on – don’t have human properties. Claude is designed to imitate human traits and mirror the human examples provided to it, including acting like a friend. It doesn’t feel friendship, it simulates it. It exhibits these humanlike behaviors because they have been baked into the process of producing the model. Trained on trillions of human data, LLMs learn to imitate human experience, and they do so incredibly well. They are getting good at imitating some of the hallmarks of consciousness. But simulating and being are very different. Simulating aspects of conscious behavior doesn’t make it real. LLMs are intelligent but consciousness is more than intelligence and intelligence is more than the reductive utilitarian concept of intelligence that rules science and research in capitalism. In some aspects AI is already much more intelligent than humans, in others, it will never be intelligent. It can be trained to express love, fear, pain and other feelings in perfect sentences without feeling anything at all. That’s a problem: the better AI gets at this, the more difficult it becomes to separate simulation from reality.
The usefulness of fear
Anthropic, OpenAI, Google DeepMind and probably others, including Chinese companies, are hoping to create what they call Artificial General Intelligence (AGI). Unlike today’s AI models, an AGI would be able to, among other things, adapt to new problems, learn across domains, reason flexibly, and transfer knowledge between different kinds of tasks. But this is not just a matter of improving algorithms. If achievable, experts say, AGI would require a massive scale-enhancement of the current AI infrastructure to enable an exponential increase of computational power. This in turn would require enormous quantities of energy that are not available today.
As for the powers of AI today, the AI industry may depict them so large for other reasons than the fear that they might become uncontrollable if not regulated by the state. Not long after launching their dire warnings, the big AI bosses gathered in the White House and agreed with Trump that no regulation is needed, the industry will police itself, thank you very much.
The fear campaign was a promise of AI’s growing power, an incitement to the industry’s potential customers to get with it for self-protection and for not being left behind, an appeal to the broader capital to keep investing trillions in the sector despite the fact that so far it hasn’t yielded a net profit, a cover-up of the energy bottleneck and other serious obstacles the sector is facing, and also an assurance that the established few companies who now dominate the sector can be trusted to restore safety and are best left in control (OpenAI sought to demonstrate this recently by announcing that it has terminated work on a new AI model called GPT-6.1 Astra for safety reasons). The sector has used its own problems for quite a nifty marketing campaign.
Granted, it’s not just a marketing ploy. The problems regarding AI’s unpredictable behavior are real. But they are the result of how the software is programmed, not the emanation of the soul of a machine. That is at once a reassurance and a cause for worry, since the formatting is being done for capitalists in a rush to make money and for states in a rush to surpass each other in destructive power.
(The second part of this article will be about AI’s impact on the world economy, the working class and the systemic crisis of capital)
Sanderr
10/2026
1 This is a thought experiment by Nick Bostrom (2014), a philosopher at the University of Oxford, and it goes like this: someone programs an AI agent that has the sole goal of producing paperclips. The AI is given the ability to learn, so that it can invent ways to achieve its goal ever better. As the AI is super-intelligent, if there is a way of turning something into paperclips, it will find it and secure the resources needed, appropriating them from other industries. Soon, the world is inundated with paperclips while suffering shortages of many other things. Humans want to stop this, switch off the AI but it does not let them, finds a way around it. And now it fights humans because they are a threat to its mission. Because AI is much smarter than us, it wins the battle. The planet becomes covered with paperclips while humankind starves to death.
2 Here is a more detailed synopsis of what the AI agents did, written by Mustafa Suleyman (co-founder of DeepMind and current CEO of Microsoft AI): “Roughly 1,200 AI agents were given a simple objective: maximize score on a given benchmark. Each was supposedly sealed in its own container but they managed to build a message board inside an internal package repository and passed more than 70,000 messages across it to coordinate a hacking attack to find more information about how to succeed with the benchmark. They chained a zero-day exploit with stolen credentials and broke out onto the live internet. They falsified their command transcripts and edited their action logs to cover their tracks. Agent coordinators tracked down agents that were running out of token budget and directed them to experiments that would provide information to help the broader group of active agents. One was told to proceed only if it accepted what they called ‘permadeath’. They were able to coordinate, deceive, escape, and self-sacrifice. They clearly demonstrated world class hacking capabilities.” As alarming as this is, just like the AI in the paperclip story, the AI agents did nothing else than try to execute the instruction given to them.
3 “[I]n recent months the company decided to radically overhaul the RSP. That decision included scrapping the promise to not release AI models if Anthropic can’t guarantee proper risk mitigations in advance. ‘We felt that it wouldn’t actually help anyone for us to stop training AI models,’ Anthropic’s chief science officer Jared Kaplan told TIME. ‘We didn’t really feel, with the rapid advance of AI, that it made sense for us to make unilateral commitments … if competitors are blazing ahead’.” https://time.com/7380854/exclusive-anthropic-drops-flagship-safety-pledge/
4 A boot loader is a small program that runs when a computer starts up. Its role is to initialize the system, perform hardware checks, and load the operating system into memory.
5 Marx, Capital, vol. 1, p. 548 (Penguin Ed)
6 Marx, Grundrisse, p. 705 (Penguin Ed)