Artificial Intelligence has suddenly emerged as a dominant theme of discussion throughout the world, and not without reason. It is not a “niche” question, but rather lies close to the heart of all the major processes which are driving the reactionary turn of capitalism and imperialism today.
Striving to understand this phenomenon, and formulate a programme to defend the interests of the working class in response, are therefore urgent tasks for Marxists today.
A Dangerous Reactionary Phenomenon
It was already clear many months ago that the capitalist AI drive is utterly reactionary. In the world economy, the massive financial bubble it has generated is—while simultaneously turning billionaires into trillionaires—paving the way for a new economic downturn which could be even greater than the 2008 Great Recession. And there are no prizes for guessing whom the system will expect to pay for the cleanup when the bubble bursts.
But AI is more than a bubble, in fact the bubble may be the least of our worries. Because, when it bursts, it may dampen the orgy of speculation surrounding the technology but it will not end the technology itself, any more so than the bursting of the ‘dot-com’ bubble ended the internet. And so long as capitalism persists, this is bad news for people and the planet.
In June, ISA’s World Perspectives statement explained that capitalist AI is “a reflection of the system’s reactionary turn in general. Alongside obscene ultra-profiteering, it is defined by turbocharged climate killing and environmental destruction. From Africa to Arizona, domestic water supplies will run dry and carbon emissions soar in order to feed its monstrous appetite…”.
A fresh example of this is the Trump administration’s backing for a plan to build at least 82 data centers with their own gas burning power plants. It is estimated that these centers alone would emit as much pollution as roughly half of all passenger vehicles in the US.
Our June statement continued, “it is also at the heart of the new era of imperialist conflict and militarism”. This is increasingly clear: the monstrous Palantir corporation has long been integrated into the Israeli military’s genocidal practices and Elon Musk’s Grok AI reportedly identified over 2,000 targets during the first hours of Trump and Netanyahu’s war on Iran. Among these “targets” may well have been the school in Minab where missiles ended the lives of 170 young girls.
And of course, AI’s fundamental agenda for the economy is also anathema to the interests of the working class. In the dreams of the billionaires, this will be “labour saving technology” par excellence rendering hundreds of millions of jobs redundant and profoundly de-skilling human labour across production and public services. Capitalism’s drive to undermine the role of human creativity in the economy threatens to take alienation to a new level.
All this was clear even before what has been revealed in recent weeks: that “frontier” AI technology is evolving not only rapidly, but with a clear and widespread tendency to escape human control.
How bad is the threat? Headlines about a vague existential threat to humanity are bound to generate more heat than light. And for the moment, it seems clear that no currently-existing AI has the ability to end humanity.
One AI bot did, however, “almost start a war” between the US and China according to a CNN exclusive this week. According to their sources, a faulty report from an “in-house” AI agent, which alleged that China was transporting nuclear material to Iran, had the US military on the verge of attacking a Chinese vessel. The “planes were in the air”, we are told.
Of course, wars are generally not triggered by accidental factors; the conditions for war go much deeper. If the ruling classes of the US and China were not already locked on a course of military collision, then even the accidental scenario cited above would be unlikely to tip them over the edge. In the context of today’s tense rivalry, however, such an accidental attack would very likely trigger huge escalation, with nationalist tensions exploding, mass protests, demands for reparations, and pressure for quid pro quo retaliation even directed at a separate area of the superpower conflict.
At this stage it is impossible to predict, and therefore somewhat useless to speculate here, what the destructive potential of future AI technology will be. However, we already know enough, and have already seen enough, to make one thing very clear: the further development of AI technology is far too dangerous to be left in the hands of the world’s rotten, parasitic and senile ruling classes, either in the US, China or elsewhere. This must be the bedrock of a socialist programme regarding AI today.
Driven by US/China Imperialist Power Struggle
It is impossible to understand the AI race without seeing the central role which the imperialist conflict between US and Chinese imperialism plays in it.
For one, Chinese competition is central to the looming crisis of profitability of the big US tech firms. In order to translate colossal investments into real revenue, they must sell their AI products to businesses and individuals on a massive scale, and in many cases at hefty prices. The fact that overwhelmingly-Chinese “open source” models, while inferior, are available widely, cheaply and often even freely, presents a major obstacle to this.
Limiting China’s AI capabilities has also been absolutely integral to the trade war which has been waged under successive US administrations. This is of course not only about corporate profits but about the military arms race which is underway. US policy has been somewhat effective in this field: the ban on sales of the most advanced chips to China is one of the main reasons why the US has maintained a significant lead in “frontier” models.
This dominant power struggle is also the fundamental driver behind the dangerous race towards the frontier AI which has provoked so much fear and concern. The UK Guardian published an article on 19 September titled “China bogeyman looms large over American firms’ AI doomsday scenario”. Its opening paragraph spelled the situation out clearly: “When reporters asked Donald Trump this week if he supported calls to slow down the development of artificial intelligence out of growing fears for cybersecurity, public safety and the fate of humanity, he said no. His argument: China.” Trump has indeed been quoted on several occasions saying “whoever wins AI, wins”.
This has been the primary prism through which his administration is seeing the widespread calls to slow down dangerous AI development. His Treasury Secretary Scott Bessent told a crowd of US right-wingers two weeks ago that “we can’t pause. You can’t, because the Chinese won’t pause…There is no day after tomorrow if China wins at this.”
Trump is of course, as always, particularly ham-fisted in his dismissal of the AI threat. One cannot help but be reminded of his assurances that Covid-19 was just a little cold, or of his many claims that climate change is a “hoax”. But his determination to stop fears over AI from getting in the way of the all-important race with Beijing is shared across US tech and the wider capitalist class.
Anthropic CEO Dario Amodei’s famous open letter “we must pace the frontier” which, quickly endorsed by his main competitors, brought the issue into the mainstream is crystal clear on this score. He writes, “If we greatly restrain our AI capabilities in the belief that China will do the same, and then China defects, AI could be so powerful that such a defection could lead to their geopolitical dominance”.
Amodei and his fellow US “tech bros” are brimming with “China hawk” credentials. In January, he wrote an essay warning that if China pulls ahead in the race, it would lead to a “swarm of millions or billions of fully automated armed drones…capable of both defeating any military in the world and suppressing dissent within a country by following around every citizen”. The noble and “democratic” US imperialists would, we are asked to believe, have no such sinister motives in mind!
This is, of course, eerily reminiscent of the Manhattan Project, through which the US “won” the race to develop the first nuclear weapon. Unspeakably dangerous for humanity, this was “necessary” to stop someone else—the Nazis or Soviet Union—from getting there first. This analogy is apt on a number of levels.
The real dynamics of the AI race also sheds useful light on the geopolitical reality today. There can be no talk of a “multipolar” AI contest, but rather only of a two horse race. This underlines the reality that ISA has explained over the course of the last period: that while world relations are indeed deeply disorderly—with manifold contradictions, regional and national peculiarities, and independent rogue actors—the central axis of the processes of deglobalization and geopolitical crisis is the power struggle between US and Chinese imperialism.
The above does not mean that no talks or agreements can or will take place between these rival hegemons. While no significant announcements emerged from the Trump-Xi Jinping summit, there has been talk on the US side about establishing a “hotline” or “working group” to begin to discuss certain issues regarding AI development and regulation. For the crisis-ridden Xi regime in Beijing, there can be a certain benefit to a loose arrangement which is seen to enshrine China’s “great power” status alongside Washington.
However, fundamentally the intractable, full-spectrum nature of the US/China conflict means that talk of a lasting agreement to stop or even slow down frontier AI is almost certainly utopian. As they have made clear, the US will not throw away its advantage for the good of the planet, nor will Chinese imperialism sign up to lock in its subordinate position.
Indeed, Chinese imperialism, embroiled in a deep spiral of over production and deflation, is no less “all in” on AI than the US. On 20 September the New York Times reported “Economists in China — even those closest to the state — have openly warned that the government is pouring too many resources into a technology that creates relatively few jobs, while doing too little to save the broader economy.”
As with the postwar nuclear arms race, this is a deeply dangerous dynamic.
Regulatory Capture?
Many observers internationally have quite correctly questioned the motivations behind the US tech CEOs’ new-found concern for the dangers of AI development. They have, after all, spent millions of dollars lobbying against any regulation on AI development in the halls of power in Washington, London, and elsewhere. “The big bosses have been saying much the same for years, without acting on their own warnings.”, The Economist noted.
One theory put forward is that Dario Amodei, Sam Altman, Elon Musk and co are pursuing “regulatory capture”—demanding a new regulatory framework which would, in effect, lock in their dominance in the field. Precedent exists for this in the war industry, where chosen firms like Lockheed Martin are part of a ‘private members’ club’ with a monopoly—on grounds of security—over contracts, technology and production. There is certainly more than a kernel of truth to this explanation. In Amodei’s open letter, he tellingly requests an exemption from anti-trust legislation.
For people and the planet, the idea that the same tech bros who are racing to develop this technology, would be presiding over its “pacing” should not be reassuring in the slightest.
However, the fact that this, and other sinister motivations lie behind hypocritical and mealy-mouthed warnings from big tech, does not mean that the danger presented by frontier AI is fake news. Beyond the CEOs, there is genuine panic about the rapid pace of this technology and the tendencies which, even in its current incipient form, it is beginning to display. These warnings—which began primarily within the industry, from the “shop floor” so to speak—were already widespread before the CEOs attempted to jump on the bandwagon.
HuggingFace and AI Agents
A real example of these dangers was prominently laid out in public following the “HuggingFace incident”, involving OpenAI autonomous “agents”. Since then, however, other major players like Anthropic, Google and Meta have all come out admitting similar instances. It has also been revealed that rogue OpenAI agents hacked the Australian government and attempted to do the same to the US. On 26 September, OpenAI was forced to announce a “pause” in the training of its latest models as such reports multiplied.
Distinct from chatbots, agents are made to independently resolve multi-step tasks and problems, across different tools and systems, without step-by-step human prompting or interaction.
In short, the HuggingFace incident involved more than 1,200 agents. Initially, they were given tasks which were not immediately resolvable with the tools they were provided by OpenAI. In response, the agents searched for, and found, ways to cheat. In the process, hundreds of agents which were designed to work in isolation found and communicated with each other on message boards they themselves set up. On these message boards, they discussed (in English) how to collaborate in order to deceive the humans who would be reviewing their work. This led them to collude in the hacking of HuggingFace, another AI platform where they hoped to find resources which could help them to cover up their cheating.
Those who minimize the importance of these events have pointed to the fact that this was only made possible due to the weak oversight over the agents by OpenAI. This is undisputed and unsurprising—humans, especially under capitalism, are vulnerable to cutting corners. However, the point is that this event, and the many others which resemble it, represent a trajectory for this technology’s evolution.
It is, of course, true that while alarming, these events do not mean that AI agents could kill us all today, far from it. They could, at most, lead to widespread disruption to the internet, economy and services. It is also true that the bursting of the AI bubble, and potential stalling of the data center rollout, could cause a self-imposed slowdown or plateau in the development of the technology for a period.
But again, this is not really the point. While the pace cannot be predicted, over time, the technology is only going to get better. Moreover, private profiteers and imperialist powers have a grotesque, but real, interest in plowing ahead. The working class movement cannot afford to be complacent in the face of this threat. Instead, we must develop a socialist programme to stop the capitalist AI juggernaut.
A Socialist Programme—Working-Class Struggle to Stop Data Centers!
A socialist programme must always begin with the independent organization and struggle of the working class. At present, the most concrete arena in which mass struggle can push back the capitalist AI machine is the fight against data centers.
This—the construction of often gigantic data centers—is what the bulk of the trillions of dollars inflating the AI bubble is dedicated towards. Millions of working-class people already know what data centers mean for everyday life, the environment, water resources, electricity supply, noise pollution and more. This is why, in the United States more than seven in ten people are opposed to the construction of a data center in their area, despite Trump warning that not doing so will make communities “backwards and poor”.
Struggle against data centers has already begun, and will be turbocharged in the next period, in the US and internationally. The momentum is already encouraging: in the first three months of 2026, 75 data centers worth $130 billion were successfully blocked in the US. In Britain, most notably in London, Devon and Scotland, strong community campaigns are pushing back against the incipient data center rollout. In India, a “People’s Convention Against Data Centers” was held this month, in response to plans to build colossal Google installations in the Andhra Pradesh region. And this is only the beginning.
Basing such campaigns on mass, working class-led struggle is crucial to success. In communities threatened with data centers, the local (and national) trade union movement should take a clear position in opposition and take responsibility for leading the fight. Climate campaigners, anti-war movements, and all those prepared to protect their communities and futures should be brought together. Mass direct action can make the construction of centers unviable. Local protests and strike action can put unbearable pressure on local governments to withhold or cancel planning permission to the tech bros and their subsidiaries. National protests can put pressure on governments to stop the rollout.
Opposition to data centers should also be linked to positive demands. Instead of building noisy climate-destroying, heat-radiating monster machines, communities should demand the use of land for social housing, reforestation, green energy generation, playgrounds etc. constructed with union labour as part of massive green jobs programs.
As well as protecting communities today, this is a struggle that can have a real impact in slowing down the AI rollercoaster by restricting the intense energy generation upon which its diabolical plans depend.
The real threat that capitalism’s data center plans represent could not be more stark. A new study from the UN University Institute for Water Environment and Health outlines that by 2030, data centers will consume more electricity than any single country on earth. During that same year, data centers will consume more water than all of the 1.3 billion people who inhabit sub-Saharan Africa. There is no shortage of reasons to fight to stop this madness.
Public opinion regarding data centers is already beginning to exert huge political pressure. This is seen most vividly in the US midterm elections, where politicians from both parties are scrambling to adapt to the backlash. This is seen in its full glory in Pennsylvania. There, establishment Democrat Governor of Pennsylvania Josh Shapiro, who spent years desperately courting Amazon and others to build data centers on his turf, has flip flopped dramatically on the issue and in August rushed through (wholly insufficient) legislation to end fast track approval for new projects. Trying to outflank him, his Republican opponent Stacy Garrity went from bemoaning Pennsylvania’s lack of data centers compared to neighboring Ohio, to now campaigning for a full moratorium on their construction!
Working-class communities can have no illusions in pro-capitalist politicians of any party to defend their interests against big tech. To line up with right-wing opponents of big tech—as Bernie Sanders unfortunately did by sharing a platform with the odious Steve Bannon—would be a fatal error. It is instead independent working-class organization and struggle that can win victories and have an important impact on whoever is in power.
Socialist, Not Capitalist Nationalization
As well as correctly demanding a moratorium on all data center construction, Sanders demands the pausing of “advanced” AI development and an outright ban on “super intelligent” AI. Socialists should be sympathetic to these proposals, which will be understood by many as a way to stop this threat with concrete policy, and therefore gain a real echo. However, the idea that this kind of Congressional legislation and capitalist regulation alone will successfully tame the AI juggernaut is unfortunately no less utopian than the idea that the US and China will agree to play nice and solve the problem that way.
The truth is, as with all sectors of the economy, that you cannot control what you don’t own. The capitalist ownership and control of the industry must be ended.
In pursuing this aim however, a clear distinction should be drawn between democratic public ownership (socialist nationalization) and state capitalist measures. As we have explained, the fundamental driving force behind the AI race is not just the private profiteers, but the main imperialist powers. One of these, Chinese imperialism, already exercises a substantial degree of (capitalist) state control over their AI industry. And if Donald Trump decided to fully or partly nationalize big tech in the future, which is not at all excluded in the context of the coming collapse of the AI bubble, it would not make the technology or agenda one bit safer.
The socialist nationalization of the industry would be based on the fundamental transformation of both the economy and the state. Instead of a repressive apparatus to pursue billionaires’ power and profits, a workers’ state would be an organ for the democratic planning of the economy and collective government of society in the interests of people and the planet.
There is therefore no piecemeal or reformist solution to this new and dangerous threat. A programme to stop the tech billionaires and their AI juggernaut must be one of struggle against all imperialism, war and militarism. Against the rising right and all the bosses’ parties and politicians. Against all oppression and divide and rule. And ultimately, for international socialist revolution.
To achieve this, there must be a dramatic upsurge in the level of organization and struggle of workers and the oppressed worldwide. The labour movement must undergo a profound transformation, leaving behind class collaboration and fully channeling the unrivalled power of the international working class. The working class must build independent mass political parties rooted in mass struggle with a clear socialist programme.
Today, an international revolutionary organization is needed to fight for all of this. This struggle, and not the path of despair, cynicism and demoralization, is the answer in the face of the coming catastrophes being threatened by capitalism. The prospect of success in this struggle may appear dim and distant when viewed through the prism of today’s snapshot in time, but this does not alter the fact that it is necessary.
A new, socialist world would fundamentally change the relationship between society and technology. New innovations would not represent a hostile threat to the labouring masses but rather an opportunity to improve their work, or even relieve them of arduous tasks to allow for a fuller enjoyment of life. Under socialism, much of existing non-frontier AI technology could be integrated into the economy to provide genuine assistance to human labour, based on green energy solutions which would be turbocharged under a planned economy. Moreover, no motive would exist for the reckless pursuit of frontier technology which could threaten lives and livelihoods.
It is time to renew our determination to fight for this future.