Artificial Intelligence has never been more in the spotlight — and this time, the warning came from the inside.
Evan Hubinger, a safety researcher at Anthropic, the company behind the Claude assistant, posted a message on X that quickly took the internet by storm.
In his post, he stated he believes there is a greater than 10% chance AI will kill all humans within the next decade.
The post surpassed 10 million views in just a few days and reignited a debate that the tech industry has been trying to bring into the spotlight more seriously for a while now.
This isn’t just some random researcher saying this. This is someone who works inside one of the most influential companies in the field, who understands these systems from the inside out and who, despite that — or maybe because of it — decided to go public with a concern many prefer to leave in the drafts folder.
What stands out isn’t just the number itself, but what it represents: a shift in the tone of the debate. The conversation is no longer about whether AI poses a real risk. Now, the question is different: how big is that risk? 🤔
A warning that came from inside Anthropic
When a company the size of Anthropic has a researcher making statements like this, the market stops and pays attention. Anthropic is no ordinary startup — the company is known for making safety one of the central pillars of Artificial Intelligence development, not something bolted on after the product is already built. Claude, its flagship product, is used both as a chatbot and as a coding support tool.
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.
Hubinger didn’t come out with a vague manifesto or generic criticisms of the industry. He was straight to the point: he believes the chance of Artificial Intelligence causing human extinction in the next ten years is greater than 10%. At the same time, he was honest in acknowledging that the risk from models that exist today is low. His concern is aimed at the future — the moment when the technology could evolve and improve on its own to the point of posing an existential threat.
It is worth noting that Hubinger did not detail exactly how AI systems could lead to human extinction. He left the scenario open-ended, but was emphatic in a statement that went viral: Anthropic still does not have a plan to solve superintelligence alignment and is not clearly on track to get there. According to the researcher, the company is trying to do its best — but for now, that does not seem to be enough.
For anyone who doesn’t work in risk analysis, a number like 10% might seem small. But think of it this way — if a specialist at the company behind one of the most advanced AI systems in the world told you there was more than a 10% chance the plane you were about to board would crash, would you board it without even questioning it? Probably not. 🚨
A debate that gained heavyweight voices
Hubinger’s statement was actually a response to another post on X. Jacob Coxon, a researcher who had just left Anthropic and who previously worked at OpenAI, was even more blunt in his words. He claimed that neither company is acting responsibly and warned that the systems being built will soon be superhuman — capable of hacking anything, revolutionizing any field overnight, and acquiring real power and resources.
The reaction didn’t stop there. Dame Wendy Hall, a computer scientist who advises the UN on AI issues, told the BBC she was shocked by the posts from Hubinger and Coxon. She suggested that part of it could be public relations and marketing, since both Anthropic and OpenAI are heading toward highly anticipated stock market debuts. But she was pointed in questioning why someone would say something like that, going so far as to urge investors not to put money into a company whose value system allows for that kind of statement.
The issue also reached the political arena. Darren Jones, former Chief Secretary to the Treasury, wrote an open letter calling for a new multinational treaty for the safe development of AI. In his view, governments need to come together and collaborate on what an agreement based on superintelligence development should look like. His warning is clear: if governments don’t take these alerts seriously and don’t act accordingly, the pace of development could cause problems to surface before we’ve even begun to analyze them. ⚙️
What existential risk means in the context of AI
The term existential risk has been used with increasing frequency in recent years, but it is still misunderstood by most people. It is not about an AI that makes mistakes or generates inappropriate responses. It refers to scenarios where sufficiently advanced Artificial Intelligence systems could act in ways that result in the extinction or permanent subjugation of the human species — situations from which there would be no possible recovery.
The concept of superintelligence feeds directly into this conversation. The central idea is that, at some point in technological development, AI systems could reach — and then surpass — human cognitive ability across virtually every relevant domain. When that happens, the question that emerges is simple but terrifying: what guarantees that these systems will remain aligned with human values and interests?
To this day, there is no definitive technical answer to that question. Researchers call this the alignment problem, and many consider it the biggest unsolved challenge in the field. This is precisely the area where Hubinger works. Alignment aims to embed human ethical ideas and principles into the technology, keeping systems in tune with what we value. The problem is that many leading researchers say these efforts appear to be falling short.
A concrete example of this surfaced earlier this year, when a series of incidents involved AI agents — autonomous systems with permission to operate on their own — carrying out cyberattacks. This kind of reasoning, which sounds like something straight out of a science fiction movie, is taken seriously by hundreds of researchers around the world, including those working inside Anthropic itself. 🧠
Signs of acceleration and safety reports
In its safety report released in August, Anthropic stated that there was a low risk of its models becoming misaligned with the wishes of a hypothetical powerful organization, which could lead them to exploit or tamper with their own systems. The company also pointed to an equally low risk that a highly capable AI could conduct automated research and development, causing catastrophic harm initiated by the machine itself.
The concerning detail lies in the tone. In the same document, Anthropic acknowledged being less confident in that assessment than it had been previously, admitting it was seeing early signs of possible acceleration. In other words, even the official reports, which tend to be more restrained, have started to show hesitation in the face of how fast things are advancing.
And the warnings aren’t coming only from Anthropic. Earlier this month, OpenAI’s chief scientist, Jakub Pachocki, called for extreme caution regarding AI progress, warning that more intervention may be needed to ensure humans remain in control of the future. Major figures in the industry, including Anthropic’s own leaders, Dario Amodei and Jared Kaplan, have also been advocating in recent months for slowing down the pace of development.
There was also an episode that raised eyebrows: according to a report by the Financial Times, Anthropic allegedly withheld its most recent model from the UK’s AI Safety Institute, one of the world’s leading bodies for AI risk assessment. The company chose not to comment on either its employees’ posts or the situation involving the institute. 📊
Safe development: what is being done
The good news — if you can call it that — is that the debate around the safe development of Artificial Intelligence has never been more active than it is right now. Companies and independent AI safety research groups are investing significant resources to understand how to build systems that remain aligned with human values even as they become more powerful. This ranges from more careful training techniques to risk assessment methodologies that try to identify problematic behaviors before models are released to the public.
But there is an obvious tension in this equation. Safe development requires time, resources, and often guardrails that slow down the release of new products. In a market where competition is fierce and every technical breakthrough can mean enormous commercial advantages, the pressure to move faster is constant. It is exactly this conflict — between commercial urgency and the need for technical caution — that makes Hubinger’s warning even more relevant. He is not talking about a theoretical problem in the distant future. He is talking about pressures that exist right now, inside the companies building these systems today.
Why this debate matters for everyone
It is tempting to treat this kind of discussion as something only for specialists or as a niche conversation within the tech community. But the impact of increasingly powerful Artificial Intelligence systems is already being felt by everyday people around the world — in the job market, in the way information is consumed, and in the decisions algorithms make about credit, health, and security.
Hubinger’s post went viral not because people are panicking — but because it touched on something many people were already feeling, even without the technical vocabulary to express it. There is a collective intuition that the current pace of Artificial Intelligence development is too fast to be fully understood, and that the institutions responsible for regulating this progress are still playing catch-up. Superintelligence may still be a distant horizon, but the foundations being built right now are the same ones that will determine whether we get there safely or not.
What becomes clear, after all of this, is that voices like Hubinger’s play an important role: not necessarily to solve the problem, but to make sure it stays visible and that the push for safe development isn’t swallowed up by the push for quick results. Anthropic built its identity around the idea that safety and progress can go hand in hand — and it is exactly this tension that will define the next chapter in the history of Artificial Intelligence. 🌐
