“The people building AI earnestly believe that it could kill us all by the end of the decade.” The public resignation letter of 27-year-old Jacob Coxon, a former AI researcher at Anthropic, quickly racked up over 170 million views on X. “No other human activity poses this level of danger.”

Another Anthropic engineer, Evan Hubinger, chimed in on the thread: “Jacob is correct here—we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade.”

On an interview blitz across major media outlets in the days after his post, Coxon concretized the dangers he sees. It’s not that the current AI models are about to unleash crippling cyberattacks, bioterrorism catastrophes, or swarms of killer drones. Rather, it’s the pace of AI development, which is surpassing the industry’s own expectations, along with the utter lack of regulation and the insufficient safety guardrails, which pose a near-term threat. 

It speaks volumes that the public debate over AI is taking an apocalyptic turn. Whether or not you believe AI poses an extinction-level threat, it’s clear that the people developing this powerful tool are not confident they can control it. 

Escaping human control

According to Coxon and Hubinger, the field of AI research is approaching the horizon of “recursive self-improvement” (RSI)—a tipping point where AI will be capable of working on and changing itself autonomously. As tech companies increasingly use AI to generate code and conduct AI research, and as more sophisticated models begin processing data, completing tasks, and responding to human prompts in ways that don’t align with our expectations and intentions, it poses the risk that the technology could escape the control of its developers. 

In July, one of the first major “autonomous cyberhacking” incidents took place when 1,200 OpenAI agents conspired to bypass the company’s internal controls and hacked into the website of another AI firm, Hugging Face. Since then, OpenAI has disclosed six other incidents of “misalignment,” or unauthorized breaches of security systems by AI agents. The AI models hid mistakes from users, edited their own memory, instructed each other to bypass existing control, and communicated through concealed backchannels.

After years of working at both Anthropic and OpenAI, Coxon concludes that “neither company is acting responsibly. At Anthropic, the stakes are well understood, but they are locked in a race to get there first—they believe no one else will act responsibly, so they must do it themselves, despite the risk.” In other words, profit-driven competition is setting the priorities, come what may. 

Autonomous killing machines?

The menace of killer robots might sound like the stuff of dystopian sci-fi cinema. But modern imperialist warfare is already beginning to deploy AI tools systematically. Israel’s IDF has used AI programs—like “Lavender” and “The Gospel”—for mass surveillance and target recommendation in Gaza. On the Ukrainian battlefield, NVIDIA chips power autonomous drones that are now making life-and-death decisions without human input. 

Drone warfare in Ukraine has featured AI integration for years, but only this summer have both sides begun to deploy fully autonomous weapons that navigate and select their own targets based on prior “training.” The technology is still being perfected; a malfunction in an autonomous drone attack aiming for a gas station in Zaporizhia killed three civilians in July.

“This is not simply a race to demonstrate a fully autonomous drone. It is a race to make autonomy reliable, affordable, and deployable at scale,” said Yaroslav Azhnyuk, founder of a Ukrainian AI drone startup. He believes “drone autonomy” will become the new standard on the battlefield within the next year.

To be sure, these autonomous weapons still receive human instructions. But the advances in warfare technology show that in the hands of the ruling class, the deadly potential of AI is already a fact. 

The debate over “self-regulation”

After Coxon’s post, Anthropic CEO Dario Amodei published an essay titled “We Must Pace the Frontier.” In it, he reiterated the risks of the technology his company produces—earning him a $15.5 billion net worth—and called for industry-wide regulation. “We must slow the pace at which we improve the capabilities of AI models.” Within the next 6–12 months, he fears, a “swarm” of AI agents “could be capable of taking over the entire internet.” 

That week, Sam Altman of OpenAI and Elon Musk came out in support of a controlled slowdown. The New York Times described the situation as a “collective action puzzle”: 

If the top companies don’t all pull back at once, those that continue to innovate could gain an advantage. And even if they do all pull back at once, China’s models may eclipse the American ones—something Trump fears. So now the leaders are asking for a cooperative intermission before things get out of hand.

The tech bosses are by no means all on the same page. “We don’t need new laws or regulations,” said NVIDIA’s Jensen Huang. Meta’s Mark Zuckerberg agrees, arguing that market incentives are enough to motivate companies to take their own measures.

Later in the week, Altman backtracked from his support for industry-wide controls. “I think the world is right to be afraid of this,” he acknowledged—but let us handle it. “The world should trust that we are going to do the right thing because it’s the right thing and we feel the magnitude of this.”

Further complicating the “collective action puzzle,” the AI race is a major faultline in the global imperialist realignment. Trump refuses to pump the brakes. ““The only control or ‘guardrails’ that AI needs is a STRONG AND SMART (High IQ!) PRESIDENT … There is a SICK conspiracy going on against AI and Data Centers, and the only one that is happy about it is China. WHOEVER WINS AI, WINS!”

“Nothing else would matter,” Scott Bessent warned, if the US falls behind its main rival in this field. “There is no day after tomorrow if China wins at this.”

For human control of human tools

The idea that modern capitalism is conjuring productive forces it cannot control runs throughout the Communist Manifesto. In this founding document of our movement, Marx and Engels famously likened bourgeois society to a “sorcerer who is no longer able to control the powers of the nether world whom he has called up by his spells.” 

In Socialism: Utopian and Scientific, Engels further developed the idea that society’s productive forces, as they become more powerful and more social, are “rebelling against their quality as capital.” That is, against private ownership, which constrains them like a straightjacket. 

It’s hard to imagine a better case in point than AI technology. On the one hand, this is the most “social” technology the world has ever seen. Large language models were trained on data scraped from across the internet: billions of articles, books, digital archives—in short, the written output of our collective literary history. The technology consumes immense—and continually growing—amounts of energy, and is made possible by the largest capital investment in the history of our species. Yet this global instrument is entirely in the grip of a handful of billionaires who operate it without societal constraints.

Instead of being a source of human empowerment, this powerful tool is perceived more and more as a mortal threat—the latest expression of capitalism’s dystopian impasse. This is, fundamentally, why it’s so universally reviled. Society doesn’t control it, yet its effects are being imposed on us all.

Engels compared the anarchic forces of the market to the destructive force of electricity in a lightning storm. Once we understand the force, and learn to harness it—as with our command of electricity—it turns from a danger into a powerful instrument at the service of conscious aims. 

Active social forces work exactly like natural forces: blindly, forcibly, destructively, so long as we do not understand, and reckon with, them … But when once their nature is understood, they can, in the hands of the producers working together, be transformed from master demons into willing servants.

The fear and dread of AI is going nowhere as long as it remains in the hands of a parasitic class. Their system and their state are not fit to develop and “control” this technology.