The big-business press and the tech billionaires talk about artificial intelligence as a machine that might someday escape human control. Human control is already being surrendered. Capitalists and the Pentagon are delegating more decisions to machines, one step at a time, under the pressure of competition. Karl Marx laid out the law that drives this more than 150 years ago.
The coercive law of competition
Every capitalist faces rivals. The one who produces cheaper and faster can take the market from the others. Marx wrote in Capital that competition makes the laws of capitalist production felt by each individual capitalist “as external coercive laws.”
The capitalist is not free to ignore that compulsion. If a rival cuts costs by speeding up production, the others have to match or beat it. If machinery allows one company to produce more cheaply with fewer workers, its competitors have to respond or risk losing market share, being bought out or going under.
Each capitalist decides how to respond. None controls the total outcome. Added together, their actions produce consequences no one planned: economic crises, industries concentrated in fewer hands and whole regions ruined.
Machinery is one of capital’s chief weapons in this fight. Marx showed that capital introduces machines to raise productivity, cheapen labor and tighten its command over workers. The Communist Manifesto described the worker as “an appendage of the machine.” The machine sets the pace, and the worker keeps up or gets thrown out.
Artificial intelligence extends machinery into supervision and decision-making. Scheduling, monitoring, hiring, firing and directing workers are as old as capitalism. Marx showed that this work of command is a function of capital, carried out for the owners by paid managers and overseers.
Now software can schedule workers, measure their performance, flag them for management and decide which cases receive human attention. In war, similar systems sort intelligence, identify and rank possible targets and recommend courses of action.
A company that can make these decisions faster and with fewer workers gains an advantage. Its rivals are driven to follow. Each step transfers another part of the decision process from people to machinery. No capitalist decides the total result. It emerges from the competition among them.
Capitalism is the runaway power
Marx described capital as self-expanding value, driven to turn money into more money without end. No individual capitalist steers capitalism as a whole. Each must accumulate or risk being pushed aside by those who do. The result confronts workers and capitalists alike as an external power.
When people describe AI as a power slipping out of human hands, they are describing a real loss of human control. But the driving force is not an independent machine will. It is capitalism, increasingly built into the machines.
An AI system can acquire enormous technical power without acquiring a social interest of its own. It can act without immediate human instructions, produce consequences its designers did not expect and become difficult to interrupt. None of that means the machine has become a new social class or an independent historical force.
The purposes still come from the capitalists who own and deploy it.
At work
Amazon already routes much of its warehouse workers’ dealings with human resources through automated systems. In February, warehouse worker April Watson suffered a concussion and was told by her doctor to work at a slower pace. It took more than a month to obtain the accommodation she needed because Amazon’s AI assistant failed to provide the correct medical form and she could not readily reach a human HR worker. Meanwhile, the system flagged her work, and she was later reprimanded for moving too slowly.
A survey by the worker advocacy group United for Respect of more than 200 Amazon and Walmart workers found that more than half said AI had already led to staffing cuts, and nearly two-thirds said their biggest worry was HR decisions being handed to automated systems.
In October 2025, Amazon announced about 14,000 corporate job cuts while pouring money into AI and stripping out layers of management.
The boss’s command over labor is being built into software.
In war
The same drive for speed appears in war. The war drive and competition among contractors reward systems that shorten the time between finding a possible target and attacking it.
The U.S. war on Iran, launched Feb. 28, 2026, showed how far this has already gone. The Pentagon used Palantir’s Maven Smart System to fuse satellite imagery, drone feeds, radar, intercepted communications and other intelligence. Maven identified and prioritized possible targets, supplied coordinates, proposed courses of action and helped match targets with weapons. Anthropic’s Claude was incorporated into the system.
The Pentagon says U.S. forces carried out more than 13,500 strikes during the first 38 days of major combat. Maven helped the military strike more than 1,000 targets in the first 24 hours.
Speed is one of the system’s purposes. The faster the target cycle becomes, the less time remains for independent human checking before the next decision.
At the beginning of the war, War Secretary Pete Hegseth promised there would be “no stupid rules of engagement.”
On the first day of the war, a U.S. strike hit the Shajareh Tayyebeh primary school in Minab, killing at least 168 people, including 120 children. In September, a U.N. fact-finding mission said there were reasonable grounds to believe the U.S. had committed war crimes in attacks on the school and another civilian facility, and said the U.S. had failed adequately to verify its targets.
A “human in the loop” does not by itself solve that problem. Human approval cannot make bad targeting data accurate. Someone still has to check what the building is, whether it is actually being used by the military and whether civilians are there.
The Pentagon’s rules
The Pentagon also uses its purchasing power to push the companies supplying these systems toward broader military use.
Anthropic had refused to remove two restrictions from its military agreement: no mass domestic surveillance and no fully autonomous weapons. On Feb. 27, Hegseth announced that he was directing the Department of War to designate Anthropic a “supply chain risk.” The designation was formally confirmed in early March and applies to use of Claude on Department of War contracts. The order came with a six-month phase-out, and Claude was still running inside Maven when the bombing began the next day. Months later, it was still in use.
The next day, OpenAI announced its own classified-network agreement with the Pentagon. Its contract permits use for “all lawful purposes.”
The Pentagon punished one supplier for its restrictions and rewarded a rival that accepted broader terms. That is the coercive law of competition at the top of the tech industry: the Pentagon uses its purchasing power to push suppliers toward unrestricted military use, and any company that holds back risks watching a rival take the contract.
The same drive runs through Washington’s policy of containing China, which some call the New Cold War. It is a form of warfare against China. Washington bans exports of advanced chips to China, blacklists Chinese firms and pours public money into U.S. tech monopolies. Every call for restraint at home gets the same answer: China will not wait. The war against China becomes the argument for handing more and more decisions to machines.
The doomsayers
Former Anthropic researcher Jacob Coxon has become one of the most prominent voices warning that AI could escape human control and wipe out humanity. His resignation from Anthropic and his warning that the people building AI believe it could kill everyone by the end of the decade went viral. Other AI executives and researchers, including Anthropic CEO Dario Amodei, have issued similar warnings.
They locate the danger in the machine because its real source — competition between capitalists and the drive of U.S. imperialism — is something they cannot picture ending. It is the line Fredric Jameson made famous: “It is easier to imagine the end of the world than to imagine the end of capitalism.”
Their remedies follow from that. They want safeguards built into the machine. A perfectly obedient machine is still a weapon when it obeys the Pentagon and still a speedup when it obeys Amazon. Safeguards inside the machine leave competition untouched, and competition is what drives the capitalists to hand control to machines.
Lasting control means ending capitalist competition
This does not mean that limits on AI are useless under capitalism. Workers are already winning them.
Hollywood writers and actors struck in 2023, followed by a nearly year-long strike by video game performers in 2024 and 2025, to win contractual restrictions on AI. East and Gulf Coast dockworkers struck in 2024 with automation as one of the central issues in their fight.
ProPublica workers staged a 24-hour strike in April 2026 after management resisted contractual protections over AI, job security and wages. They later won a first contract that included requirements for transparency around AI use.
At five McClatchy newspapers in the Pacific Northwest, journalists struck for a day in May and later won contract language barring management from using AI to replace staff, reduce their pay or cut their working hours.
The companies knew what their products would do to news workers. Court documents unsealed on Sept. 17, 2026, in the copyright suit brought by the New York Times and other publishers against OpenAI and Microsoft show Microsoft staff debating whether training AI on millions of news articles was the “largest theft of labor in human history.” In 2024, the head of OpenAI’s ChatGPT team wrote that AI products would substitute more and more for publishers’ work as they improved. Neither company stopped. In a competitive market, the one that holds back hands the field to a rival.
These strikes show that workers can force limits on how the bosses use the technology now.
But every such limit confronts the same underlying pressure. A capitalist who can cut labor costs, accelerate production or gain an advantage through automation has an incentive to do it. The pressure returns as long as capitalist production is driven by private profit and competition.
Full social control over these machines therefore means more than better safeguards. It means taking the tech monopolies out of private hands and planning the development and use of AI around human needs, including reducing the working day rather than eliminating workers’ livelihoods.
The class capable of imposing that control is also the class that builds, operates and maintains the machinery.
Competition is handing more work and more war to machines. Workers can fight for limits now. Ending the force that continually drives the process requires taking control of the machines out of the capitalists’ hands.
Join the Struggle-La Lucha Telegram channel