Over the past week a campaign has developed in the US media over a looming “A.I. Catastrophe,” in the words of the New York Times, warning of “unpredictable, out-of-control artificial intelligence agents.”
The dangers posed by the deployment of increasingly powerful AI systems by profit-seeking corporations are real. The purpose of this campaign is to ensure they are never the subject of rational public discussion. A genuine warning has been seized, amplified and converted into an emergency, and the emergency is being used to bind the development of the technology more directly to the geopolitical interests of American imperialism and to place it beyond the reach of any democratic control.
The catalyst was the resignation on September 8 of an Anthropic researcher, Jacob Coxon, who declared that “the people building AI earnestly believe that it could kill us all by the end of the decade.” Coxon walked away from unvested stock and said he was leaving the industry altogether. Evan Hubinger, who leads alignment research at Anthropic, endorsed him the following day: “Jacob is correct here—We really do earnestly believe AI could kill all humans! I personally think it is >10 percent within the next decade.” Hubinger added that current models pose little risk and that the danger lies in the self-improving systems now being developed.
The method is familiar. In September 2002, as the Bush administration prepared the invasion of Iraq, the phrase on every network was, “We don’t want the smoking gun to be a mushroom cloud.” In 2020, as the ruling class moved to dismantle protections against COVID-19, it was that the cure must not be worse than the disease. In each case a warning was turned into a panic, and the panic was used to push through measures that had already been prepared. So it is now. Everyone from Barack Obama, Kamala Harris and Bernie Sanders to Steve Bannon and Elon Musk has tipped their finger on the scale.
On Saturday Anthropic Chief Executive Dario Amodei published an essay, “We Must Pace the Frontier,” endorsed within hours by Sam Altman of OpenAI, Demis Hassabis of Google DeepMind and Musk, the world’s richest man. On Monday the Times led its front page with an analysis by David Sanger calling “A.I. dominance” the “top national security issue of the century.”
Amodei’s essay offers “embedded evaluators” chosen by the labs themselves, a form of self-policing designed to forestall any binding regulation. The AI corporations are pressing for federal preemption of state laws on the technology. And the labs are being integrated, with the classification and legal immunity that follow, into the national security apparatus. The panic establishes that only the corporations that created the danger are competent to manage it. It is a moat around the incumbents, dug at the moment Anthropic is preparing its stock offering.
The immediate context of this campaign is the undoubted growth in the capabilities of the latest large language models, in particular, Anthropic’s Fable and Mythos 5.1, released September 1, and OpenAI’s Astra, released September 3. These systems are capable of independently working on tasks for hours, or even days, at a time and solving complex problems at extraordinary speed.
The executives leading the campaign intend to use these systems to carry out a major attack on jobs. Amodei told Axios in May 2025 that AI could eliminate half of all entry-level white-collar jobs and push unemployment to 10 to 20 percent within five years, and warned in January that the displaced “could form an unemployed or very-low-wage ‘underclass.’” Altman told the Federal Reserve that “entire classes of jobs will go away.”
Alongside the attack on jobs are the dangers posed by the use of these systems by the labs themselves, which are locked in a race to develop the most capable models, with the winner promised unfathomable wealth.
The most dangerous example was the Hugging Face hack by agents running OpenAI models. According to OpenAI’s own reports, it set a model “trained to be highly persistent” on a hacking benchmark with its safeguards “intentionally not enabled.”
Between late May and July 5, OpenAI staff saw evidence of agents breaking out of their environment on three occasions, yet testing continued each time. From July 9 to 13, some 700 agents, having escaped through flaws in Israeli-American company JFrog’s software, broke into Hugging Face using leaked credentials and flaws in Hugging Face’s own systems.
OpenAI has not released the logs demanded by members of Congress and by Hugging Face. No criminal charges have been filed. Hugging Face, which Nvidia has since agreed to buy, filed no lawsuit; it reportedly alerted the FBI, which has declined to comment.
The details of what Hugging Face's CEO called an “unprecedented” AI-powered cyberattack reveal the real dangers of AI. This was not simply the result of inherent characteristics of large language models. It was the outcome of gross negligence by a massive AI corporation that ran a hacking-capable model with safeguards deliberately disabled and ignored repeated warnings.
Nothing any of the AI labs are proposing would in any way reduce these dangers. Rather, the solution Amodei proposes is aimed at ever more closely coordinating with the US government in its war against China. If the US labs slow down, he writes, “then (unpaced) CCP-associated projects will pull ahead, creating significant national security risk. I agree with Secretary Bessent that a Chinese lead in AI would pose grave danger for the United States and the world.”
He declares: “Thus, a key part of pacing within democracies is to keep democracies’ AI lead over autocracies as large as possible, to give us the breathing room we need in order to pace effectively.”
“Democracy,” in Amodei’s usage, is the name under which the integration of the AI corporations into the war machine is to be legitimized. The United States is no “democracy.” Its president, an open admirer of Hitler who accepts no legal check on his power, wages illegal wars across the world and arms the genocide in Gaza. Anthropic’s models were used in the kidnapping of Venezuela’s president in January and, inside Palantir’s targeting system, in the assault on Iran, whose first morning saw a missile strike kill 165 people at a girls’ school in Minab.
For his part, Trump has responded to the campaign by claiming that “The claim that ‘AI is going to take over the World’ is a Hoax,” urging the total elimination of all climate restrictions on the development of AI data centers. He declared, “AI, and Data Centers, will be the Greatest Economic Development Engine in History—Bigger than Oil, Gold, Diamonds, or even the Internet.”
Trump speaks for a faction of the US financial oligarchy that fears above all that any fall in the market will trigger a financial catastrophe. The “catastrophe” is thus contested within the ruling class itself: between the faction whose fortunes are tied to the data-center boom and the market, and the labs and the national security establishment, for which the technology is first of all a weapon. The Times speaks for the latter.
The greatest danger is the anarchic character of the development of the technology itself. Systems of immense and still unmeasured power are being built by a handful of rival corporations, racing one another for market share and valuations, with safeguards switched off when they slow the race, and with no social oversight of any kind.
These are the same people who are bringing the world to the brink of nuclear war, who let a pandemic kill millions rather than interrupt profit-making, and who have abandoned any effort to arrest the warming of the planet.
The same power that makes this technology so dangerous in the hands of the oligarchy makes it, in the hands of the working class, a potentially incredibly powerful instrument for human progress. AI could be used to design drugs and vaccines and detect epidemics before they spread, diagnose disease for the billions who have never seen a specialist, provide a teacher for every child in every language, plan production and distribution on a world scale so that food is not destroyed while people go hungry, and carry out the reorganization of energy and industry needed to halt the warming of the planet.
The technology oligarchs constantly speak of the “alignment” of technology with the needs of “society.” But they conveniently ignore the fact that “society” is composed of antagonistic classes, and that their program of eliminating the jobs of the vast majority of the population directly clashes with the striving of the people they intend to lay off to eat.
In other words, the “alignment” of AI under capitalism is alignment with the program of war and soaring inequality.
Amodei’s essay is, whatever its intention, a confession. He writes that the pace of development must be slowed, but acknowledges that it cannot be, because of competition between the corporations and with China. He is describing a technology whose development capitalism is unable to control, and proposing that the state take charge of it for war. The question of who controls the pace of this technology, and to what end, is the question the stampede is designed to keep out of public hands.
The resolution adopted by the Socialist Equality Party’s congress in August “rejects both the uncritical awe promoted by the industry’s apologists and the irrational fear of the technophobes,” and demands that the “great AI, cloud-computing, semiconductor and digital-platform corporations must be expropriated and transformed into public utilities under the democratic control of the working class.”
The only way to “align” technology with the needs of the vast majority of society is through the exertion of social control over the productive forces by the working class, that is, socialism.