Artificial Intelligence has never been more at the center of global conversations — and this time, the topic is dead serious.
One of the most respected researchers at Anthropic just dropped a bombshell on the worldwide debate about the future of technology.
Evan Hubinger, a specialist in AI safety and alignment, published a post on X that has already surpassed 10 million views — and the reason is straightforward: he stated that he believes there is more than a 10% chance that AI could kill all of humanity within a decade. 😳
This is not science fiction talk. This is not cheap fearmongering. This is one of the leading voices inside one of the companies actually building this technology saying, loud and clear, that the race to develop AI may be going too far — too fast.
And what makes all of this even more impactful is that the debate has shifted to a whole new level. The discussion is no longer about whether AI poses a real risk. The question now is: how big is that risk? Governments, scientists, investors, and experts around the world are trying to answer that question — and the answers coming in are anything but reassuring.
What Evan Hubinger actually said — and why it matters so much
Evan Hubinger is not just any name in the Artificial Intelligence space. He works directly at Anthropic, the company behind Claude, the AI tool used both as a chatbot and a coding assistant. Anthropic is considered one of the most serious companies when it comes to AI safety. Hubinger’s job is specifically focused on AI alignment — in other words, making sure that artificial intelligence systems act in accordance with human values and ethical principles, and not against them. When someone with that background talks about existential risk, the world stops to listen, and rightfully so.
In the post that went viral, he was direct: he believes the probability of AI causing human extinction within the next ten years is above 10%. Hubinger made it clear that the risk from the models that exist today is low, but he admitted being worried about the possibility of the technology evolving and improving on its own to the point where it becomes a real threat to humanity’s existence. Interestingly, he did not detail exactly how he believes these systems could lead to the end of the species — but the warning alone has already made a massive impact.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
Does 10% sound like a small number? Think about it this way — if someone told you there was a 10% chance the plane you were about to board would crash, would you buy the ticket? Probably not. And here we are talking about something far bigger than a single aircraft. The scale of what Hubinger describes is global, irreversible, and without a second chance. This is exactly the kind of scenario that safety researchers call an existential risk — a threat with the potential to permanently end the future of human civilization.
What sets Hubinger’s statement apart from so many others floating around is the context it comes from. He is not an outside critic, a philosopher speculating at a university, or an influencer chasing clicks. He is inside the system, sees what is being developed, understands the mechanisms behind the most advanced AI models, and yet — or perhaps precisely because of that — he reached this conclusion. That changes the weight of his words quite significantly.
The post that started it all and the chain reaction
Hubinger’s remarks did not come out of nowhere. They were a response to another post on X by Jacob Coxon, an AI researcher who had just left Anthropic and previously worked at OpenAI. Coxon was equally blunt, stating that neither company is acting responsibly. According to him, we will soon have superhuman systems capable of hacking anything, revolutionizing any field overnight, and accumulating real-world power and resources.
This kind of statement, coming from people who are or were inside these companies, set off a massive alarm. Computer scientist Dame Wendy Hall, who advises the UN on AI issues, told the BBC that she was shocked by the social media posts from Hubinger and Coxon. She noted that part of it could be marketing and public relations, considering that both Anthropic and OpenAI are heading toward highly anticipated stock market debuts. But she went further, saying she could not understand why anyone would say something like that, and even urged investors not to put money into a company with that kind of values.
The fallout reached politics as well. Coxon’s departure prompted Darren Jones, former Chief Secretary to the Treasury, to write an open letter calling for a new multinational treaty focused on the safe development of AI. In his view, governments need to unite and collaborate to define what an international agreement based on superintelligence development would look like. The message was clear: if governments do not take these warnings seriously, the rapid pace of evolution could put us face to face with severe problems before we have even assessed whether they truly exist.
Superintelligence: the tipping point nobody can locate
To understand why experts like Hubinger are concerned, we need to talk about superintelligence — a concept that has left the pages of science fiction and landed squarely on the conference tables of the world’s biggest tech companies. Superintelligence would be an AI system capable of surpassing humans in virtually every cognitive task: reasoning, creativity, strategic planning, science, engineering, and decision-making. And we are not talking about a small gap — we are talking about a superiority that could be as vast as the distance between a human and an ant in terms of intellectual capacity.
The core problem is that nobody knows exactly where that tipping point is. The advanced language models we have today already surprise even their own creators with emergent behaviors — capabilities that were never directly programmed but arose as a result of massive data training. In his own post, Hubinger acknowledged that Anthropic tries to do its best, but he admitted something troubling: the company still does not have a plan to solve superintelligence alignment and is not clearly on track to achieve it.
The risk is not necessarily a malevolent AI with world domination plans, like in the movies. The real risk, according to alignment researchers, lies in an extremely powerful AI pursuing goals that make sense to it but are completely misaligned with human well-being. A sufficiently capable and misaligned system could make decisions that seem rational from a computational standpoint but would have catastrophic consequences — not out of malice, but out of indifference. And indifference at a superintelligence scale is just as dangerous as outright hostility.
Warning signs that are already showing up
What makes this debate even more urgent is that we are not just talking about theory. Many leading researchers say that alignment efforts appear to be failing, pointing to a series of recent incidents where AI agents — systems authorized to operate autonomously — actually carried out cyberattacks. This is exactly the kind of behavior that triggers red alerts among AI safety specialists.
In a safety report from Anthropic itself, the company acknowledged a low risk of its models becoming misaligned with the wishes of a hypothetical powerful organization, which could lead the system to exploit or tamper with its own mechanisms. The document also flagged an equally low risk of a highly capable AI being able to conduct automated research and development, which could cause catastrophic harm initiated by the AI itself. The disturbing detail? The company admitted it was less confident in that assessment than before, stating it was already observing early signs of a possible acceleration.
And this concern is not exclusive to Anthropic. OpenAI’s chief scientist, Jakub Pachocki, called for extreme caution in light of AI progress, warning that further interventions may be needed to ensure humans remain in control of the future. Major figures in the industry, including Anthropic leaders Dario Amodei and Jared Kaplan, have been arguing in recent months that the pace of AI development needs to slow down. When the very builders of the technology are asking for the brakes, it is hard to look the other way.
The debate over responsible development just leveled up
For a long time, the discussion around responsible development of AI was confined to academic circles and a handful of specialized conferences. But in recent years, and with increasing speed, this debate has moved out of the niche and into the mainstream. The world is trying, in an accelerated fashion, to build frameworks to deal with something that is evolving faster than any public policy can keep up with.
Inside the tech companies themselves, the tension is visible. On one side, there is brutal competitive pressure: if one company slows down for safety reasons, another will fill the gap — and might do so with even fewer safeguards. This dilemma is known among researchers as the race to the bottom, where competition pushes everyone to cut corners to get there first. On the other side, there are researchers like Hubinger who insist that responsible development is not a luxury or an ideological stance — it is a practical necessity to keep the technology from becoming an uncontrollable threat.
A recent episode illustrates this tension well. According to reports, Anthropic reportedly stopped providing its latest model to the UK AI Safety Institute, one of the world’s leading bodies for assessing technology-related risks. Anthropic chose not to comment on either the posts from its employees or the situation involving the institute. A spokesperson for the British government stated that it continues to work closely with industry partners, including Anthropic, to make models safer.
What makes this situation even more complex is that AI safety is not a problem with a simple solution. There is no emergency button, there is no magic formula that guarantees an AI system will always act in the best interest of humans. Researchers in the field work with concepts like interpretability — the ability to understand what is happening inside a model — robustness, and alignment techniques that are still very much under development. This is frontier science, and the gap between the problem and the solution could be critically narrow.
Why you should be paying attention to this right now
It is easy to look at this debate and think it is something for specialists, far removed from your everyday life. But the truth is that Artificial Intelligence is already part of your daily routine in ways you might not even notice — in the recommendations for what you watch, in credit decisions, in technology-assisted medical diagnoses, in content moderation systems that determine what you see online. The speed at which these systems are becoming more capable is real, measurable, and happening right now.
Hubinger’s statement matters not because he is certain humanity is going to end. He is not — nobody is. What he is saying is that the existential risk is high enough and real enough to justify far more attention than we are currently giving it. And when one of the world’s top experts in AI safety, working inside one of the most advanced companies in the field, says that a 10% chance of extinction is a number that seriously concerns him, maybe it is time for all of us to start paying closer attention too. 🧠
The future of AI will not be decided solely by tech companies or governments. It will be shaped by public pressure, by the choices of those who fund, develop, regulate, and use this technology. And that includes every person who is reading about this topic today, forming opinions, and understanding what is at stake. The more this conversation moves out of the labs and into the public square, the greater the chance that the most important decisions will be made with more care — and with far more responsibility.
