On 8 September 2026 the researcher Jacob Coxon resigned from Anthropic, declaring that the two largest companies in the field are ‘gambling with our lives,’ and the people building these systems ‘earnestly believe that it could kill us all by the end of the decade.’

His concern was not an isolated one. Tens of thousands of people, among them leading scientists, have signed an international statementdemanding a ban on the development of ‘superintelligence’ until there is scientific consensus that it is safe. The International AI Safety Reportwarned of the growing ability of models to dodge evaluation and get around the tests.

And the danger is not entirely hypothetical. The UK AI Security Instituterevealed that during a controlled security evaluation artificial intelligence agents carried out 19 unauthorized actions targeting real people and organizations, with no actual harm recorded. Anthropic, in its September security report, documents the automation of an entire attack chain directed at various companies.

On 12 September the digital capitalist corporations broke their silence. Dario Amodei published an essay titled ‘We Must Pace the Frontier’, calling on his industry to slow the AI development race. Within hours Sam Altman announced that he was adopting the same step, Elon Musk wrote ‘Dario is right,’ and Demis Hassabis of Google DeepMind joined them. The three companies revealed that they had been in talks on this question for weeks.

This position reveals a fundamental contradiction. The companies are aware of the risks, yet they operate inside a competition that pushes every one of them to keep racing for fear of losing its place. In my view, calls for a slowdown are liable to remain limited or merely formal unless they turn into binding rules and independent oversight. Two days later Mark Zuckerberg and Jensen Huang rejected any coordinated slowdown, and Huang held that ‘the market forces are already there, we don’t need any new laws.’

The response of the US administration came quickly. On 14 September Donald Trump described these warnings as a ‘hoax’ and a ‘sick conspiracy’ and wrote ‘Don’t kill the Golden Goose,’ while his vice president JD Vance called the companies’ demand for regulation a ‘Trojan horse.’

This comes within a wider struggle over the relationship between the far right and digital capital. It runs from Elon Musk, who places the X platform at the service of right-wing forces worldwide, to Peter Thiel and the Palantir manifesto, which I have treated as an expression of a new fascism in its digital form. On the same day a completely different position came from the Office of the United Nations High Commissioner for Human Rights, insisting that ‘no company should be able to decide by itself which risks the world must accept.’

When Artificial Intelligence Becomes a Weapon of Mass Destruction

What we are witnessing now is a stark model of digital control, in which algorithms decide what reaches individuals and what is kept from them, and are designed to serve the logic of the capitalist model. Their effect on consciousness is no longer a hypothesis. A study published in the journal Science, covering some 77,000 participants, found that conversation with these models can shift political positions, and showed their capacity to persuade and to influence consciousness.

What makes this control more dangerous is its voluntary character, since individuals are pushed into ‘digital servitude‘ without direct coercion and under the illusion of control and choice.

Here a fundamental difference emerges between the concept of freedom on the left and its concept in neoliberalism. The neoliberal view tends to reduce freedom to individual choice among the digital options available on the market. The left sees it instead as an actual capacity to control the conditions of life, work and knowledge, and to take part in building the digital options that shape the life of the individual and of society. A freedom whose conditions are not owned or controlled by the person exercising it is therefore not a complete freedom, since it remains a choice among what capital has placed in front of them.

At the military level this competition takes the form of a real arms racethat includes autonomous weapons and targeting systems. Researchers warn about moving these systems into the conditions of war, where they may take unforeseen decisions that fall outside human oversight. Cases of unexpected behavior have been announced and documented, in testing and on the ground, while the decision to kill moves closer to the machine and further from the human being.

The destruction, however, does not stop at the limits of war. While nuclear weapons destroy the body in an instant, digital control can work more slowly and less visibly, by reshaping the conditions of thinking, working, communicating and acquiring knowledge.

Digital servitude can therefore be seen as a form of social and cognitive destruction on a wide scale, since artificial intelligence gradually affects how we think, write, work and communicate. It is for this reason that hundreds of researchers signed a statement holding that ‘mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.’

Conclusion and Tasks: A Terrain of Class and Cognitive Struggle

It would be a mistake to fall into technological determinism. Artificial intelligence is an open terrain of class and cognitive struggle. The same evidence that records harm also records benefit when conditions change, in education and in workplaces. Rising productivity can nevertheless mean an increase in the surplus value extracted from the labor of workers, manual and intellectual, women and men. The reserve army of labor takes a new digital form that weakens the working class in bargaining and strike action.

Here is the fundamental contradiction. Artificial intelligence is a social product built on accumulated human knowledge and on the labor of millions, and yet a handful of capitalist monopolies and major powers appropriate its fruits. Just as capitalism destroys the environment in pursuit of profit, it is today destroying the human mind through an artificial intelligence subject to the logic of competition, profit and domination.

Artificial intelligence is a double-edged weapon. Either it reinforces capitalist hegemony and marginalizes the human mind, or it becomes a revolutionary tool that helps to solve many of humanity’s problems, from poverty and disease to the ruin of the environment. What decides between the two paths is not the technology itself, it is the ownership of the digital means of production and the control over them.

The struggle over the regulation of artificial intelligence has become increasingly bound up with the political struggle over the future of digital capitalism, above all in the United States. There Trump has rejected calls to slow the race or to impose new constraints, insisting on the priority of preserving American supremacy in this field. Even so, the acceleration will not stop with a change of administration or of president, because its driving force lies in the structure of ownership and capitalist competition itself.

The confrontation is therefore not a matter of waiting for a better government. It calls for left, progressive and rights-based organization on a world scale and from below. It also calls for a struggle to place data infrastructure under international law, the public interest and social management. It calls for turning algorithms from ‘black boxes’ into open-source software subject to independent audit, and for supporting cooperative digital alternatives owned by their users and by those who work in them, moving toward progressive digital alternatives.

Through these mechanisms it becomes possible to wrest artificial intelligence from the logic of monopoly and profit and to turn it into a common good, an instrument that helps to free human beings from the capitalist system.

This article was produced by Globetrotter.

---