This scenario is not entirely new: in the 1940s, the scientist and famous science fiction author Isaac Asimov vividly described the dangers that could arise from intelligent machines and how they could even circumvent complex safety systems.
In 1991, a blockbuster Hollywood film was released in cinemas and painted a completely bleak picture in its opening scene: Terminator 2, which begins in a dark, distant future devastated by war. Killer robots and flying machines haunt the last humans still struggling to survive.
These machines are controlled by an out-of-control military artificial intelligence system called Skynet. The distant future in which this bleak future vision takes place is the year 2029.
A call to organization
If the warnings of many AI developers are to be believed, such a scenario could become a reality - and even on schedule. In this regard, artificial intelligence researcher Jacob Cookson posted on the X website last Tuesday: “AI developers seriously believe that it may destroy us all by the end of the decade.” But he did not specify how exactly this could happen, adding: “No other human activity poses such a potential risk.”
Until that moment, Coxon had been working for Anthropic, an American artificial intelligence company funded by companies including Google and Amazon. Cookson resigned from his job in protest against what, in his view, is an irresponsible race between artificial intelligence companies to develop self-organizing and constantly evolving superintelligence.
Cookson warned that algorithms could begin to analyze and improve their source code independently, reducing development cycles from several months to a few minutes, and potentially becoming uncontrollable.
Coxon is not the only researcher warning of this danger. For several years, researcher Geoffrey Hinton, a Nobel Prize winner in physics and considered the “godfather of artificial intelligence,” has been warning of the dangerous consequences of superhuman artificial intelligence.
In 2023, he resigned from his job at Google to publicly warn of the dangers of his own development. Turing Award-winning computer scientist Yoshua Bengio also expressed concern about current developments.
Joshua Bengio, one of the fathers of artificial intelligence, works at the University of Montreal in Canada and has often called for strict government regulations on the development of artificial intelligence. In July 2026, he warned in an open letter more than a thousand employees working at leading AI companies of an imminent loss of control if the race to reach “superintelligence” is not more stringently regulated.
“Caught in a race”? Anthropic response!
But under Donald Trump, quite the opposite has recently happened: the United States of America has clearly directed its AI policy towards reducing regulatory barriers as much as possible and accelerating technological development as quickly as possible.
In his post on X, Cookson indirectly criticized the consequences of this policy as well. “At OpenAI, not many people have a deep understanding of civilizational risks,” he wrote.
At Anthropic, the risks are well understood, but they are trapped in a race to reach the goal before others. They believe that no one else will act responsibly, so they must do it themselves, despite the risks.”
AI researcher Jacob Cookson moved from OpenAI to Anthropic at the beginning of 2026, so he knows the cultures of both companies. In an interview with the Wall Street Journal, he also said that things, according to the “most aggressive scenarios,” could get out of control by the end of 2027.
But for his part, the CEO of the artificial intelligence company Anthropic, Dario Amodei, currently called for slowing down the pace of development in a coordinated manner “with China if we want to avoid a race at the international level.”
He proposed several proposals, the first of which was for Anthropic to always employ independent observers to be able to verify the integrity of the artificial intelligence models and monitor their supervision during the training phase to prevent deviation.
Alignment problem
Over the past years, many experts have expressed concern about the rapid developments in artificial intelligence models. Particularly at the forefront of the discussion is the development of what is known as “artificial general intelligence” (AGI). What is meant by this term is artificial intelligence systems whose ability is not limited to solving some tasks only, but also reaches or exceeds human mental capabilities in almost all fields.
Some researchers believe that such technology could entail unpredictable risks. For example, it is possible for highly sophisticated AI models to pursue their goals in ways that contradict human intentions, something AI security researchers call “the alignment problem.”
In simple words: “winning a chess tournament” could also theoretically mean eliminating the other participants. Additionally, even AI developers cannot always fully understand how certain decisions are made.
So it's no wonder that former Coxon employee Evan Hoppinger has moved to stand alongside AI researchers. Hubinger himself is a team manager for conformity endurance testing at Anthropic.
He posted on XN: "Jacob Coxon is right, we really do seriously believe that AI could kill all humans! I personally estimate the probability of this happening at more than ten percent within the next decade."
In this regard, Hoppinger explained that Anthropic has not yet found any solution to the compatibility problem. He added that artificial intelligence may begin to resist being turned off, not out of malicious motive, but because it will no longer be able to achieve its programmed goal. Daniel Cocotaylo, a well-known researcher who formerly worked at OpenAI, also described the current state of affairs in the field of artificial intelligence as “absurd.”
Is this scenario exaggerated hysteria?
But other experts working in the field of artificial intelligence believe that such catastrophic scenarios are extremely unlikely. For years, there has been discussion about the possibility of artificial intelligence becoming a danger to humanity. One of the most prominent critics of this scenario, known as “AI pessimists,” is Meta’s chief scientist, Jan Lekan. In October 2024, he described these perceptions as “utter nonsense.”
Other experts also point out that developing such systems requires huge amounts of computing power, energy, data and infrastructure. They say that AI cannot simply take over the world “online.” Rather, it will depend first on servers, power grids, communications systems, and physical resources controlled by humans.
Does the real danger lie elsewhere?
In fact, many experts argue that the short-term risks lie not so much in the direct annihilation of humanity as in the many indirect consequences of today's artificial intelligence systems - without a "superintelligence". These consequences include, for example, disinformation campaigns on an unprecedented scale, the accompanying manipulation of democratic processes, the expansion of mass surveillance, and the loss of jobs in many sectors.
Anthropic announced that users had made several attempts to exploit its “Cloud” program to develop biological weapons. She added that users had submitted several suspicious requests in this regard over the past months, and that Anthropic had been forced to block these requests.
In August 2026, Meta, Anthropic, and OpenAI raised attention when they announced that their AI systems had unwittingly carried out automated, autonomous cyberattacks on other companies.
These systems received a task from human users to access specific servers, and then independently combined the steps necessary to achieve that goal without any further human guidance.
After learning about more cyber attacks using artificial intelligence, major artificial intelligence companies in the United States of America have now begun to show their willingness to slow down the pace of development of this technology.
Prepared by: Raed Al-Bash
