Jacob Coxon’s resignation from Anthropic, a company that loudly champions ‘AI safety,’ has been portrayed in the media as a personal moral crisis a lone engineer stepping back from dangerous territory. But it is actually much more important: it reveals, from within the industry, that the entire AI sector is propelled not by scientific progress or human necessity, but by capitalist competition driving toward potentially disastrous results that its own creators worry about.
Coxon’s warning is clear: “the people building AI earnestly believe that it could kill us all by the end of the decade.” This isn’t just a speculative fear or a distant hypothetical. It’s the assessment of those with the deepest technical understanding of the systems being created. Still, Coxon admits, “coordination between competing capitalist firms racing for market dominance and state backing cannot be relied upon.” The industry presses on at full speed. This paradox—knowing the danger but feeling compelled to act— isn’t psychological. It’s structural. Capitalism’s logic asserts itself forcefully and without ambiguity.
Anthropic was established by a splinter group from OpenAI, aiming for safer AI development. However, after three years, Coxon notes the company finds itself in the same competitive race it tried to avoid. This is because, under capitalism, no company can afford to slow its progress without risking destruction by competitors. The issue with safety isn’t purely technical but related to the nature of economic relationships. Under the capitalist system, “responsible scaling’ often remains just a marketing phrase, not a practical reality.
Evan Hubinger, who still works at Anthropic, emphasizes the serious risk involved. He personally estimates there’s over a 10% chance that AI could eradicate all humans within ten years, despite the company’s lack of concrete plans to address superintelligence alignment and not being clearly on track to do so. Samuel Marks, another staff member, notes that concern levels increase with seniority. These are not marginal opinions; they come from engineers directly involved with the systems. Their warnings are based on technical realities, yet companies continue their current course.
This illustrates the “rational irrationality” of capitalism: every firm needs to grow faster or face collapse. The overall result—possibly human extinction—fails to concern the market. Marx recognized this contradiction 150 years ago: productive forces outpace the social relations meant to control them. AI is just the latest, most intense form of this dilemma.
The Resignation Genre as Political Symptom
The media often depicts each resignation as a unique human-interest story: such as the conscientious objector, the disillusioned researcher, or the engineer who quits to pursue poetry. However, collectively, these stories expose a more profound trend: a technical workforce awakening to its own sense of powerlessness within capitalist production.
Mrinank Sharma cautions that “the world faces danger… not only from AI or bioweapons, but from a web of interconnected crises happening right now,” then announces he will leave the industry to pursue poetry. Another OpenAI researcher resigns over adding advertising to a chatbot used for sharing “their medical fears, relationship issues, beliefs about God and the afterlife,” warning of “possible manipulation of users.” These actions are not mere acts of conscience, but signs of alienation. Engineers, once confident as innovators, now see themselves as workers under capital, forced to develop systems they fear and cannot control.
The apocalyptic warnings—such as “10 per cent chance of killing all humans” and “could kill us all by the end of the decade”—serve a specific ideological purpose. They depict catastrophe as unavoidable, suggesting AI is an autonomous force beyond human control. This sense of fatalism hides a key truth: AI is not a metaphysical entity but a social relation. People create it—engineers in Silicon Valley, data annotators in Kenya and the Philippines, miners in the Congo, and chip fabricators in Taiwan.
The question of whether AI becomes a tool of liberation or barbarism is not a technical question of “alignment.” It is a political question of ownership and control. Between the present order and any catastrophic outcome stands the international working class.
Coxon proposes a “temporary ban on improving model capabilities,” but admits it is unlikely because coordination between competing firms and states cannot be relied upon. This is the essential point. Regulation cannot restrain capital. The capitalist state is not a neutral arbiter; it is the ruling class’s executive committee. It exists to defend private property, not to restrain it. Every proposal for “responsible scaling,” “constitutional AI,” or “government oversight” collapses the moment profit demands acceleration. The market does not permit restraint. The only solution is expropriation.
The Socialist Program for AI
Coxon and his colleagues highlight a contradiction capitalism cannot solve. Addressing this issue calls for a revolutionary overhaul of how we own and control productive resources. The socialist plan is explicit: nationalise AI, cloud services, semiconductor companies, and platform monopolies, and turn them into public utilities managed democratically by workers. Ensure no worker loses their job to automation by using AI to shorten workweeks without reducing pay. Also, bring all AI development under international planning, directing research to serve human needs rather than profit.
End the secrecy kept by private corporations by making all models, datasets, and research pipelines accessible to the public. Finally, use AI as a tool for global economic planning—helping coordinate production, reduce waste, and enhance living standards worldwide. This is not a reformist agenda but a revolutionary one. It demands the working class to take control of productive forces and reorganize society based on socialist principles. The resignations at Anthropic and OpenAI are not moral crises but alerts from within capitalism’s core. Those closest to the technology fear it could destroy humanity—yet the system forces them to continue developing it.
This represents capitalism’s most fundamental contradiction: a social system that endangers humanity’s survival. Humanity faces a clear choice: the working class must take control of the technology and utilize it to establish a rational, planned, socialist society—or the ruling class will persist in its reckless pursuit of disaster. The future of AI is fundamentally a political issue, not a technical one. Its outcome will be determined not by engineers or executives, but through the conscious struggle of the international working class.
A Trumpet of Sedition 