Dramatic warnings about the existential risks posed by artificial intelligence to humanity are nothing new.
As early as the 1940s, the renowned scientist and science fiction author Isaac Asimov explored potential risks and unintended consequences of intelligent machines.
In 1991, Hollywood blockbuster “Terminator 2” imagined a dystopian 2029 in which an out-of-control military AI called Skynet directs killer robots in a war against humanity.
If the warnings of several of the biggest names in AI are to be believed, a similar scenario could become a reality — even before 2029.
Growing calls for strict regulation
“The people building AI earnestly believe that it could kill us all by the end of the decade,” AI researcher Jacob Coxon posted on X last week, without exactly explaining how it might happen.
“No other human activity poses this level of danger,” he added.
Coxon also revealed that he had quit his job at Anthropic, a US-based AI company, in protest against what he saw as an irresponsible race among AI companies to develop an ever better, self-improving superintelligence.
He warned that algorithms could begin to analyze and improve their ownom several months to just a few minutes and thus make them uncontrollable
Coxon is not an isolated voice. Nobel Prize-winning physicist Geoffrey Hinton, often called the “Godfather of AI,” has been warning for years about the serious consequences of artificial superintelligence.
In 2023, he left his job at Google to publicly warn of the dangers of AI technologies he helped pioneer. Yoshua Bengio, a pioneer of artificial neural networks and deep learning, has also expressed concern about current developments and called for government regulation.
As recently as July 2026, more than 1,000 employees of leading AI companies warned in an open letter of an imminent loss of control if the race for “superintelligence” were not subject to significantly stricter regulations.
Caught in an unregulated race?
Despite the growing warnings, US President Donald Trump’s administration has clearly shifted its AI policy toward minimizing regulatory hurdles and accelerating AI development.
In his X post, Jacob Coxon indirectly criticized the consequences of this policy: “At OpenAI, many have not deeply internalized the civilizational stakes. 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.”
Coxon, who had moved from OpenAI to Anthropic in early 2026, is familiar with both corporate cultures.
In an interview with The Wall Street Journal, he said that under the “most aggressive scenarios,” things could spiral out of control as early as the end of 2027.
America’s AI boom: Who pays the price?
Anthropic CEO Dario Amodei, for his part, is currently calling for a coordinated, industry-wide slowdown in development to buy time to address the risks and prevent any misuse of AI.
In a lengthy essay shared on X over the weekend, he wrote that in six to 12 months, AI agents “could be capable of taking over the entire internet” and potentially cause “hundreds of billions of dollars in damage” without proper guardrails.
Both Elon Musk, who runs xAI, and Sam Altman, CEO of OpenAI, said they agree with Amodei.
The alignment problem
In recent years, numerous experts have voiced concern about the rapid AI advances, particularly the development of artificial general intelligence, which refers to systems that not only solve individual tasks but also match or exceed human intellectual capabilities in nearly all areas.
Sophisticated AI models, for instance, may pursue goals in ways that diverge from human intentions, something AI security researchers refer to as the “alignment problem.”
On top of that, even developers can’t always fully understand how AI makes certain decisions.
And so, it’s no surprise that Coxon’s former colleague Evan Hubinger came to his defense.
“Jacob is correct here — we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade,” Hubinger, who himself is the team lead for alignment stress tests at Anthropic, wrote on X.
In his post, Hubinger made clear that Anthropic hasn’t yet found a solution to the alignment problem and warned that an AI model might begin to resist being shut down, because that would limit its ability to achieve its programmed goal.
Exaggerated hysteria?
Other AI experts, however, consider such doomsday scenarios to be extremely unlikely.
Among the most prominent critics of the so-called AI doomers is Meta’s chief scientist, Yann LeCun. In October 2024, he dismissed such concerns as “complete nonsense.”
Still, other experts note that developing AI systems requires enormous amounts of computing power, energy, data and infrastructure.
They argue that an AI model cannot simply take over the world because it would need to rely on servers, power grids, communication systems and physical resources that are controlled by humans.
China’s humanoid robots put to the test
Does the real danger lie elsewhere?
In fact, many experts see the short-term risks not so much in the extinction of humanity as in the numerous indirect consequences of AI systems that already exist even without superintelligence.
These include disinformation campaigns on an unprecedented scale and the related manipulation of democratic processes, an expansion of mass surveillance, and job losses across numerous industries.
Just last week, Anthropic revealed that there had been several attempts by users to misuse its Claude AI models for research that could support biological weapons development. The company said it had banned the accounts involved.
In August, Meta, Anthropic and OpenAI caused a stir when they announced that their AI systems had unintentionally carried out autonomous hacking activities.
Following reports of further AI hacking attacks during testing, even top US AI frontier labs are now expressing a willingness to slow down the development of the technology.
This piece was originally published in German.
