The doomsday drum keeps beating
Jacob Coxon resigned last week. He worked for the AI company Anthropic. In a viral public letter, he declared that human extinction at the hands of artificial intelligence was now imminent. It was a shocking claim. It was also an old one. This was not the first time a serious person had raised the alarm about the existential risks of the technology they helped to create.
The doomsday drum has been beating for over a decade. Professor Stephen Hawking said it clearly back in 2014. He warned that the full development of artificial intelligence 'could spell the end of the human race'. That prediction was made a little less than a decade before the public got its hands on the generative AI inside the first version of ChatGPT. He was not alone. Other scientists joined the chorus. Tech leaders did too. From university labs to Silicon Valley boardrooms, the message has been consistent for years. A superintelligence could end humanity.
So what happened? Did the industry listen to one of the world's most famous scientists? No. The warnings from Hawking and others seem to have been treated not as a red light but as a starting pistol for a global arms race. The noise grew louder. The investment grew larger. Every company from Google to Microsoft has since piled into the field, committing billions of pounds to building the very technology that a vocal minority insists will destroy us. The development has not just continued. It has accelerated beyond all expectation.
A contradiction sits at the heart of the AI boom. One group, including some of the technology’s own creators like Jacob Coxon, claims it poses a unique threat to our species and must be controlled. Another group, which includes many of the same companies and investors, is spending unprecedented sums of money to ensure it gets built as fast as humanly possible. The fears have been shaken off. The pursuit continues. Why has a decade of dire warnings from respected voices failed so completely to apply the brakes?
An arms race measured in dollars
The answer is money. It is always money. For the chief executive of a publicly traded company, the risk of human extinction is a philosophical problem for a future generation, while the risk of seeing a rival like Microsoft steal market share is an immediate crisis that will get you fired by Tuesday. The market has no patience for hypotheticals. Boardrooms in Silicon Valley are accountable to shareholders who demand quarterly growth, not to philosophers contemplating eternity. That is the reality. This immense commercial pressure forces a kind of tunnel vision, a strategic deafness to warnings that do not relate to the next earnings call. An abstract danger cannot compete for attention with a concrete one. The fear of being left behind is more powerful. An analyst downgrade is a more terrifying prospect than a robot apocalypse.
This logic creates a vicious spiral. It is an arms race. A company that pauses for reflection on the warnings of a departing researcher like Jacob Coxon will not be praised for its caution. It will be punished by investors for its hesitation, watching its best engineers leave for competitors who are still moving at full speed while its stock price collapses. The only thing considered more dangerous than building this potentially world altering technology is the certainty of letting a competitor build it first and control it exclusively. Imagine the conversations inside Google. If Microsoft successfully integrates a new generative model into all its products, the pressure to respond with something bigger becomes overwhelming. There is no alternative. This is a commercial imperative. It pushes all other considerations into the background.
The sums involved are vast. The investment is measured in billions of pounds. This is not speculative capital being gambled on a long shot, it is a strategic necessity for the biggest corporations on earth who believe they are fighting for their very survival. Once a company has committed that level of resource, it creates a powerful momentum that is almost impossible to stop or even to steer. The prize is not just a new product category. The prize is everything. The winners believe they will own the foundational layer of the entire twenty first century global economy, a technological dominance making previous monopolies look small. This creates a simple calculation for boards in London and San Francisco, a calculation where the immediate financial obliteration from falling behind always outweighs the abstract possibility of a future catastrophe. The maths is simple. Extinction is a rounding error.
Two meanings of 'AI safety'
This confusion is profitable. The term ‘AI safety’ itself has been split into two entirely different meanings, a division that technology companies are happy to exploit. One meaning belongs to the philosophers and the departed researchers like Jacob Coxon. It is the big one. It is the Stephen Hawking meaning. This is the argument about preventing an artificial superintelligence from escaping human control and wiping out civilisation, a problem of cosmic significance that may or may not ever happen.
The other meaning of AI safety is the corporate one. It is mundane. It is about today. This version is not concerned with robot overlords or the end of the human race. It is concerned with avoiding lawsuits, managing bad publicity and preventing brand damage. Corporate safety teams are not designing ethical cages for future gods. They are building filters to stop their chatbots from generating racist instructions, producing biased hiring recommendations or plagiarising copyrighted material in a way that might attract the attention of a regulator or a very expensive lawyer. This is risk management. It is about spreadsheets and legal opinions. It is not about saving the world. It is about protecting next quarter’s earnings report from the immediate and very tangible threats that current, flawed AI models pose to the business.
This ambiguity serves the companies perfectly. It is a brilliant piece of misdirection. When executives announce they are prioritising safety, they allow politicians and the public to imagine them wrestling with the profound existential questions raised by thinkers like Hawking. This performance of deep thought makes them look responsible. It makes them seem serious. It buys them goodwill. It costs them nothing. In reality, the budget for ‘safety’ is spent on the corporate version, on the urgent, practical work of making their products less likely to cause a public relations disaster or trigger a billion pound fine. The existential debate, for all its noise and all its drama, provides a useful smokescreen, allowing the commercial race to continue at maximum speed behind a veneer of philosophical caution. They get to have it both ways.
The brakes are physical, not philosophical
The real limits on AI are not philosophical. They are physical. They are made of concrete, copper and cooling fluid. The greatest challenges to the AI industry are not being debated in university senior common rooms or at Silicon Valley summits about the nature of consciousness, but are instead being fought in drab town hall planning meetings over zoning permits, water rights and electricity grid capacity. This is the messy reality. This is where the exponential growth curve of software collides with the stubbornly linear world of building things. You cannot simply will a gigawatt of power into existence. You cannot conjure a data centre from thin air, no matter how clever your algorithm is. The bulldozers and the planning lawyers move at their own speed.
An artificial intelligence does not live in the cloud. It is a pleasing metaphor. It lives in a box. It lives in a very specific, very expensive, very hot box called a data centre. These vast, windowless warehouses are the hidden factories of the twenty first century, and their appetite for resources is ferocious. They need power. Huge amounts of it. A single large data centre can consume as much electricity as a hundred thousand homes, putting an immense strain on national and regional power grids that were already struggling to cope with the transition to electric cars and renewable energy. This is not cheap power. It is constant, 24 hour a day, industrial scale consumption that sends utility bills into the hundreds of millions of pounds and requires dedicated substations to be built, often over the objections of local communities who do not want the giant humming transformers and high voltage pylons marching across their fields.
Then there is the water. All that computational effort, all those trillions of calculations per second, generates a colossal amount of waste heat, and cooling these server farms requires millions upon millions of litres of water every single day. This is a problem. It is a big problem in areas already facing drought or water stress, where the idea of a technology company getting priority access to scarce resources over local farms or household taps becomes politically toxic. Suddenly, the global AI race that began with academic papers hits a very local and very solid wall. Planning applications for new facilities are quietly withdrawn. They are denied. Protests are organised by residents armed not with philosophical treatises but with environmental impact assessments. Local councillors who have never heard of the company Anthropic or Stephen Hawking find themselves voting on whether to approve a new pumping station or allow a tech giant to tap into a community aquifer. The theoretical end of the human race, as warned by Hawking back in 2014, feels very distant when you are worried about the village well running dry next summer. The opposition is not about ethics. It is about noise, dried up riverbeds and the strain on public resources.
Regulators are worried about copyright, not cyborgs
While prophets of doom fret about rogue AIs, governments are moving at their own speed. It is a slow speed. They are not writing laws to prevent human extinction. They are writing laws to stop copyright theft, to break up monopolies and to protect personal data. The immediate problems for AI companies are not philosophical. They are legal. They are expensive.
Legislators and regulators are focusing on tangible harms that exist right now, creating a web of compliance costs and legal risks that act as a powerful, if mundane, brake on development. Forget cyborg assassins. Think about the lawsuit filed by The New York Times against OpenAI and Microsoft, which alleges that millions of copyrighted articles were used without permission to train the models that power services like ChatGPT. This is not an abstract risk. It is a direct challenge to the core business model of generative AI, threatening billions in damages and forcing a fundamental rethink of how these systems are built. The problem is not a superintelligence deciding to wipe out its creators. The problem is a judge in a New York courtroom deciding that a company owes a fortune for stealing its training data. This creates immediate uncertainty. It demands teams of expensive lawyers.
Similar battles are being fought on other fronts. The UK’s Competition and Markets Authority is examining the tight relationships between tech giants and AI startups, probing whether these partnerships stifle competition and lock up the market before it has even matured. Regulators in the European Union are implementing the AI Act, a sprawling piece of legislation that imposes strict transparency and risk management obligations on developers, creating a compliance regime that is costly and complex to navigate. These are the real world pressures. They are the daily concerns inside boardrooms. The debates are about data protection impact assessments, not the dawning of a new mechanical consciousness. The cost of navigating privacy rules like GDPR, proving ownership of training data, and satisfying competition watchdogs is a real and growing line item on every balance sheet. It is a threat far more immediate than anything imagined by Stephen Hawking.
The argument is a sideshow
The argument over human extinction is a useful sideshow. It is loud. It is dramatic. It attracts headlines about rogue machines and the end of humanity, but it distracts from the immediate and expensive problems that actually govern the industry’s trajectory. The future of artificial intelligence will not be decided by philosophers debating the consciousness of a large language model. It will be decided by the mundane, grinding realities of resource consumption and local politics. A data centre needs power. Lots of it. It also requires immense volumes of water for cooling, a resource that is becoming increasingly scarce and politically sensitive. These are not theoretical problems. They are physical limitations that translate directly into spiralling operational costs and fierce opposition from communities who do not want a power hungry, water guzzling facility in their backyard. The fight is about utility bills.
This is the real war. It is fought not with algorithms but with planning applications, environmental impact reports and legal challenges. The people shaping AI’s development are not futurists. They are council planning officers, competition regulators and intellectual property lawyers. Their concerns are not about a hypothetical superintelligence emerging in 2045. Their concerns are about the anticompetitive effects of Microsoft’s partnership with OpenAI today, the legality of using copyrighted news articles for training data right now, and the precise compliance costs associated with the EU’s new AI Act next year. These issues create enormous financial and legal friction. They slow everything down. A company’s ambition to build the world’s most powerful model is ultimately constrained by its ability to secure a reliable power supply at a sustainable price and navigate a labyrinth of international regulations.
The apocalyptic warnings, from Stephen Hawking a decade ago to Jacob Coxon last week, make for compelling drama. They are a spectacle. But the story of AI’s evolution is being written in far less thrilling documents. It is being written in invoices from utility companies. It is found within legal briefs filed in New York courtrooms. Its future is shaped by the dense technical annexes of data protection legislation from Brussels. The limits are boring. They are expensive. The genuine brakes on this technology are the balance sheet, the courtroom and the planning committee, not a sudden attack of conscience about creating a mechanical god. The real risk is not that AI becomes too powerful. The real risk, for the companies building it, is that it becomes too expensive to run and too difficult to defend in court.
Sources. BBC News Business: Why doomsday warnings are not the only threat to the AI juggernaut. Guardian Technology: Why a decade of doomsday warnings failed to slow the AI race.
Analysis. Drafted with AI assistance from the sources listed above and reviewed by an editor before publication. Jnews links to the organisations it writes about.
