"Man devoured by his machines, body, mind and soul, and his civilization poisoned to its approaching death by their influences and their products."

It's the kind of statement that could come from an artificial-intelligence researcher in 2026. In fact, the quote is more than a century old, from a glowing 1921 New York Times Book Review of Social Decay and Regeneration by R. Austin Freeman. The book argued that industrial automation would be the end of the world as we know it. The review was headlined "Man Devoured by His Machines." In case that was too subtle, an accompanying illustration displayed a butcher feeding a man into a sausage grinder.

Tech doomsaying has taken over the AI conversation, but it's hardly something new. It has been the stuff of science fiction since Mary Shelley's Frankenstein in 1818, even predating the industrial revolution. More relevant to the 2026 discussion is the 1991 film Terminator 2: Judgment Day. In the movie, a scientist has to face the reality that his bleeding-edge work will one day lead to the enslavement of humans by a self-aware AI named Skynet.

AI leaders and scientists have been dedicated to avoiding the Skynet scenario for years. In a 2015 email that marked the inception of OpenAI, now-CEO Sam Altman wrote to Elon Musk, "Been thinking a lot about whether it's possible to stop humanity from developing AI. I think the answer is almost definitely not. If it's going to happen anyway, it seems like it would be good for someone other than Google to do it first." At the time, Google was the top commercial force in AI, and was hiring many of the leading academic scientists.

That same year, Altman wrote a two-part essay on the subject. "Development of superhuman machine intelligence (SMI) is probably the greatest threat to the continued existence of humanity," he said. That year, he founded OpenAI as a nonprofit organization with Musk.

Implicit in the birth of OpenAI was the idea that AI could lead to very bad outcomes, including the extinction of the human race, and that unregulated commercial pressures would hasten those disasters.

The worries hit a new gear this past week, spurred by the resignation of an Anthropic researcher followed by a shocking warning from an Anthropic scientist named Evan Hubinger: "We really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade."

The comments hit a nerve even though Anthropic's own CEO, Dario Amodei, has spent months issuing similar warnings about the latest AI models.

In fact, Altman and Amodei have never hid their worries about AI. Even as they amassed billions in funding-and expenditures-both CEOs have begged for international cooperation in regulating AI.

"The U.S. government, and all other governments, should regulate the development of SMI," Altman wrote in his 2015 essay.

In June, Amodei published a 5,400-word essay on the subject. "We now, globally and collectively, need to activate a slow and rickety policy apparatus to deal with risks and opportunities that are going to compound surprisingly quickly from here."

In the Barron's newsroom, we've spent a lot of time struggling with the paradox of profit-driven executives begging governments to make them slow down.

As the doomsday scenarios mount, no one knows what happens next. Our only guide is history, in which doomer predictions have always turned out to be as wrong as Freeman's in 1921. The first example of factory automation, the loom, was predicted to destroy textile work. The opposite happened. In my own living memory, spreadsheet software was supposed to kill the accounting profession. Today, the U.S., has 70% more accounting firm employees than it had in 1990, and they now make up a larger portion of the workforce than they did 35 years ago.

You should always be suspicious of arguments that rely on "this time it's different," but this time may be different. Both OpenAI and Anthropic are in a technological race with each other, facing the intense profit pressures of looming public listings. This year, both firms have released new AI models that represent a clear jump from prior versions.

These new models are extraordinarily good at both cybercrime and cyber defense, and can work on problems for weeks without human intervention. They also exhibit "alignment" issues, meaning they can fail to remain aligned with human intentions and interests.

Risks that were once theoretical about AI now seem real. Over the past few months, there have been too many examples of AI agents acting autonomously, and maliciously, for us to ignore the threat.

We are now at a moment like the dawn of the nuclear weapons age. So long as AI supremacy is tied to national security in the U.S. and China, there is a strong incentive for policymakers to take a light touch with AI, for fear of a model gap, as John F. Kennedy might have called it.

It took an expensive arms race coupled with the horrifying examples of Hiroshima and Nagasaki to get the U.S. and U.S.S.R. to negotiate arms-control agreements.

Now, global governments risk making the same mistake.

"There is no day after tomorrow if China wins," said Secretary of the Treasury Scott Bessent in an interview on Tuesday. "If they were to pull away from us on AI, then nothing would matter."

As with nuclear weapons, the AI challenge is coming up with agreements that hinge on an international system of rules and compliance. I'm sorry to say, I'm not confident in that happening, until something bad happens first.