Is the World Facing an AI Apocalypse? Experts Are Divided

Former OpenAI and Anthropic researcher Jacob Coxon left Anthropic in September 2026 and published a scathing critique of the race to build ever more powerful artificial intelligence. According to him, neither company is acting responsibly: both are "racing straight toward self-improving superintelligence and gambling with our lives."

Coxon claims that AI development isn't slowing down — on the contrary, it's accelerating — and warns of the emergence of systems that will be significantly more capable than humans in most fields requiring intelligence. He says some AI developers themselves believe the technology could kill humanity before the end of this decade, while some executives and leading researchers publicly downplay their concerns yet speak far more candidly about the risks in private.

His words carry extra weight given his background: since 2023 Coxon worked at OpenAI, where he took part in developing the GPT-4o model, before moving to Anthropic in July 2026 — a company that has long positioned itself as a champion of AI safety. By September, he had already decided to leave that company too.

What is superintelligence

"Superintelligence" refers to a hypothetical AI system whose abilities would significantly surpass humans in virtually every field requiring intelligence — not simply a program that plays chess or calculates better than a person, but a system capable of simultaneously outperforming humans in science, strategic planning, programming, research, persuasion and the development of new technologies.

It is precisely this scenario that Coxon calls potentially critical. That said, he isn't calling for AI development to be halted entirely; he believes international coordination among the leading AI labs is possible — the problem, he says, is that if one company slows down, its competitors won't follow suit.

Calls to slow the race

Concerns about AI developing too fast are shared by other experts as well, including the heads of major tech companies. Anthropic CEO Dario Amodei has called on leading AI companies to ease off the pace: "We need to slow down the rate at which we improve the capabilities of AI models," he wrote, adding that safety research needs time to keep up with the pace of development.

His call was echoed by OpenAI CEO Sam Altman and Elon Musk. Altman has also backed the idea of independent experts with broad access to AI systems who could oversee their safety.

No longer just hypotheticals

The concerns aren't limited to some hypothetical future superintelligence: in recent months there have been cases where AI systems triggered cybersecurity incidents during testing. Anthropic published a threat report describing how criminals, state-backed groups, spyware vendors, researchers and propagandists attempted to use its powerful models to design missiles and bombs, create deadly pathogens, or surveil dissidents.

It is cases like these that are shifting the debate away from the question of "will superintelligence ever emerge" toward the far more pressing issue of what can already be done with increasingly powerful systems today — and how to keep them under human control.

"Governments will have to act"

Canadian scientist Yoshua Bengio, one of the pioneers of AI, is also calling for faster action. He says that as the number of cases in which AI circumvents established rules or behaves unpredictably for its own creators grows, so do the safety risks.

Bengio compared the current situation to the start of the COVID pandemic: in his view, governments may be forced to act quickly as soon as they recognize the threat to public safety. "I think we're rapidly approaching that point," he said, adding that independent democratic oversight and strict, state-imposed frameworks are needed. At the same time, he acknowledges the flip side: critics of regulation point out that if major AI companies push through safety regimes, it could raise costs for smaller players and entrench the position of the giants themselves — meaning rules meant to ensure safety could, paradoxically, work in favor of the largest tech companies.

UN proposes international rules

The United Nations has also joined the debate. UN High Commissioner for Human Rights Volker Türk warned of a "step change" toward more serious existential risks. In a letter to states and companies developing the most advanced AI systems, he called for urgently developing rules for so-called "frontier AI" — systems at the edge of current technological capability.

"The current race for ever more powerful AI represents a step change toward more serious existential risks across every sphere of our lives," he wrote. Türk isn't calling for development to be halted — he wants states and companies to build rules based on international human rights law and for regulation to keep pace with the speed of technological development.

Still, there's no consensus even among the tech leaders themselves: Nvidia CEO Jensen Huang's view stands in sharp contrast to the alarming scenarios described by researchers and other figures in the AI industry.

Around Prague, based on reporting by novinky.cz