Photo credit: Pablo Munoz
Article by Christina Vaughan, 10th September 2026
There has been a lot of discussion today about a statement by Senior AI Safety Researcher Evan Hubinger at Anthropic, who puts the probability of AI causing human extinction at above 10%.
It’s a statistic that has stopped many of us in our tracks. Read that again: there could be a greater than one-in-ten chance that artificial intelligence will ultimately cause human extinction.
It is the sort of statement designed to make you sit up and listen. And, like many people, my immediate reaction was: Really?
AI is surrounded by increasingly polarised debate. Depending on what you read, it is either going to transform medicine, science, education and human productivity or take our jobs, destroy creativity, undermine truth and, at the most extreme end of the argument, potentially threaten human life itself.
The reality, unsurprisingly, is more nuanced.
What does that 10% actually mean?
Highly respected AI researchers take the possibility of catastrophic AI risk extremely seriously. Surveys have also found that a meaningful proportion of researchers assign a non-trivial probability to outcomes involving human extinction or permanent loss of human control.
This should concern us.
But it is important to understand what these numbers are and, importantly, what they are not.
A figure such as 10% is not an actuarial calculation based on historical evidence. Humanity has never developed artificial superintelligence before, so there is no dataset from which anyone can scientifically calculate the probability of what it might eventually do.
Hubinger has said he puts the probability of AI causing human extinction at above 10%, in the context of concerns about increasingly capable and autonomous AI systems. These figures are essentially expert judgements about uncertain future scenarios. That distinction matters, particularly when a speculative probability becomes a frightening headline such as “There is a 10% chance AI will wipe out humanity by 2030.”
We simply don't know that.
What we do know is that AI is developing extraordinarily quickly, and some of the people who understand the technology best believe that sufficiently advanced AI could create very serious risks if its capabilities develop faster than our ability to control them.
The people worrying about catastrophic AI risk aren't all cranks or sensationalists.
A major statement from the Center for AI Safety says that mitigating extinction risk from AI should be treated as a global priority alongside pandemics and nuclear war. Its signatories include Geoffrey Hinton, Yoshua Bengio, Demis Hassabis, Dario Amodei, Sam Altman, Ilya Sutskever and many other leading figures in AI.
A 2023 survey of around 2,780 researchers publishing at leading AI conferences found substantial concern about extremely bad outcomes from advanced AI. Depending on the precise wording of the question, median estimates for extinction or similarly permanent human disempowerment were around 5–10%.
That's significant.
If credible scientists think there's even a small possibility of something catastrophic, society should invest seriously in understanding and preventing it. This really deserves attention rather than dismissal.
Nobody currently knows whether we'll have the kind of autonomous, recursively improving superintelligence required by many of these scenarios by 2030. Nobody knows exactly how such a system would behave. Nobody knows what safeguards, regulation, architectures or international agreements will exist by then.
There are also extremely knowledgeable AI researchers who assign much lower probabilities to extinction scenarios and believe some of the more apocalyptic arguments depend upon assumptions about future AI capabilities that may never materialise in that form.
What actually worries me more in the near term
The risks we already know are real are considerably less cinematic.
AI can amplify fraud and cybercrime; generate extremely convincing misinformation; concentrate economic and political power; displace or radically change certain kinds of employment; enable increasingly sophisticated surveillance; reproduce human biases at enormous scale; and potentially assist people attempting dangerous biological or cyber activity.
And then there's something particularly relevant to what we are building at Cultura: the integrity of the information environment itself.
If humanity reaches a point where enormous quantities of synthetic material circulate without provenance and nobody knows what is authentic, who created something, whether the people depicted consented, or where training material came from, that's a profound societal problem even if no Terminator ever arrives.
These are risks that don't belong to the future
While the world debates hypothetical superintelligence, there are important questions about AI that are already sitting in front of us.
Where does the data used to train these systems come from?
Did the people represented in that data consent to its use?
Do the creators of the material have rights?
Can its provenance be established?
Does the dataset accurately represent the diversity of the real world?
What biases are embedded within it?
And increasingly: how will we distinguish authentic human-created material from synthetic content?
These questions may sound less dramatic than human extinction, but their consequences are already real.
AI systems learn from data. If that data contains historical bias, poor representation, questionable provenance or material used without appropriate permissions, those problems don't simply disappear. At scale, technology can amplify them.
That is one of the reasons we believe responsible dataset creation matters.
The opportunity is enormous too
It is easy for the AI conversation to become dominated by fear.
But that would be equally misleading.
Used responsibly, AI has extraordinary potential: accelerating scientific discovery, improving healthcare, expanding access to education, increasing accessibility, helping us understand complex systems and giving people entirely new creative and productive tools.
We are already seeing glimpses of that future.
The question therefore shouldn't simply be:
“Is AI good or bad?”
Technology rarely offers us such a convenient binary choice.
A better question might be:
“What kind of AI do we want to build, and what decisions should we be making now to help ensure that future?”
Humans are still in the room
This is the part of the conversation I find most important.
AI is not some distant force arriving from another planet. Human beings are building it. Human beings are choosing the data. Human beings are designing the systems, establishing the commercial models, writing the regulation and deciding what standards we are prepared to accept.
That means we still have agency.
At Cultura, our contribution to that enormous challenge is necessarily a small one, but it is something we care deeply about.
We believe AI training data should be ethically sourced, properly licensed and documented, genuinely representative and created with respect for the human beings who appear within it.
Every person in our datasets is a real, living, breathing person, not simply a data point.
They have been photographed or filmed by creative professionals. They have been paid. Appropriate releases exist. The provenance of the material is understood.
And representation matters too. If AI is going to become embedded in the systems through which billions of people experience the world, the data underpinning those systems should reflect the world in all its complexity.
So, perhaps we are asking the wrong question
Will AI wipe out humanity?
Nobody knows.
And despite some alarming headlines, nobody can responsibly tell us that there is a scientifically established 10% probability that it will.
But some very serious people believe advanced AI could eventually present catastrophic risks. We should listen to them.
At the same time, we shouldn't allow dramatic predictions about an uncertain future to distract us from the decisions being made today.
Because the future of AI isn't predetermined.
It will be shaped by researchers, governments, technology companies, creative industries, dataset providers, and ultimately by society itself.
And perhaps there is a philosophical connection here to something we talk about a lot at Cultura. AI isn't inherently the future, nor is it inherently the apocalypse. Human beings are making choices right now about what kind of technology it becomes.
The extraordinary opportunity of this moment is that we are still early enough to influence what comes next.
One thing I've learnt since beginning to work in AI dataset development is that human lives rarely unfold according to the neat strategic plans we like to construct for them. Sometimes there's a moment of faith in God, the Universe, oneself, whatever language we give it, when the rational spreadsheet doesn't quite justify the leap, but something says: this matters: let's find a way.
Perhaps there is something of that in this moment too.
For those of us working at the intersection of creativity, technology and artificial intelligence, our responsibility may not be to predict the future.
It is to help build the version of it that we want to live in.
Christina Vaughan
Founder & CEO, Cultura Creative
10th September 2026