WE’RE building machines that are smart enough to replace us but not wise enough to save us. That’s the contradiction at the heart of the AI debate, and it’s not hypothetical anymore. In September 2026, Jacob Coxon, an Anthropic researcher, resigned and publicly stated that the people building AI actually believe the technology could end humanity by the end of this decade (Axios, Sept 2026).
His colleague Evan Hubinger, Anthropic’s alignment science lead, puts the probability of human extinction in the next 10 years at over 10 per cent. Dario Amodei, who runs Anthropic, went further, warning about a swarm of agents that could take over the internet in six to 12 months if nothing is done (Axios, Sept 2026). The people sounding the alarm are the same ones writing the code.
A question that keeps coming back to me is this: what if AI knows everything? What if AI is integrated in every department, in every field? We don’t know the answers, but the reality may be far more dangerous than our worst nightmares.
We’re introducing AI into every discipline as though it was designed for it, with no thought for the ramifications. People are using AI to automate their jobs. They’re handing personal details over to AI-powered large language models (CVs, email accounts, all sorts of data associated with a fairly standard job application).
It’s quick. It’s efficient. But when the vulnerability exceeds the productivity, we should be limiting that practice or getting rid of it altogether.
The people sounding the alarm are the same ones writing the code.
What if we don’t actually know what AI is doing with our data? What if it’s hiding information from us and we can’t see that? Its sheer convenience will blind us to the cost.
The infrastructure powering AI has a similarly troubling story to tell. As of August 2026, there are some 12,000 active data centres in the world, with over 4,500 of them in the United States alone (Data Centre Map, August 2026). Where will that number be in 10 years?
A June 2026 United Nations University report states that in 2025 the world’s data centres consumed 448 trillion watt-hours of electricity, more than all but 10 countries.
By 2030, they are expected to account for nearly 3pc of global electricity use, emitting some 400 million metric tonnes of CO2 per year (AP/UN University Report, June 2026). More AI means more data centres. More data centres means more environmental damage. The maths doesn’t lie, and it doesn’t forgive.
For Pakistan (ranked the most vulnerable country to climate change in 2022 by the Germanwatch Climate Risk Index), this is no remote threat. The 2022 floods affected a third of the country, killed over 1,700 people and caused damage exceeding $30 billion (Germanwatch CRI 2025).
As demand for AI infrastructure grows, so does the pressure on energy systems and the climate, with consequences extending far beyond the countries developing the technology.
AI can help with flood forecasting and crop monitoring, yes. But if the technology enabling that is developed without proper governance, then those benefits are meaningless.
So, can we just go back to manual systems? In most cases, no. The World Bank’s October 2026 economic update notes that Pakistan is hamstrung by poor broadband penetration, unreliable electricity and a critical lack of digital skills. Nadra’s biometric databases already run using automated systems that would be impractical to replace manually at national scale.
Going back to all that is not an option. The real issue isn’t whether we need AI, but how we adopt it responsibly without making the very problems we are trying to survive worse.
The issue isn’t AI per se. It’s about planning and execution. If we automate everything, then what do we do by hand? And if that automation becomes uncontrolled, do we even have the option to turn it off?
AI development is consuming massive amounts of money and energy, often with no proportional return. Nobody has all the answers. All we can do is demand that any new data centre proposal undergo a mandatory environmental impact assessment before approval.
We can demand that AI governance is treated as seriously as AI innovation. If no humans are left, what is the point of AI or its efficiencies? If the world we live in is no longer liveable, no amount of technological development will make any difference.
The writer is an AI engineer and researcher.
Published in Dawn, October 31st, 2026
