The prospect of an AI apocalypse is “exceptionally high”, amid growing fears that computers could eventually identify humanity as a threat to be eliminated.
That warning came from Paul Christiano, an OpenAI board executive, during a stark lecture this week. He argued that autonomous AI systems could go rogue and try to seize control of the world within three years.
Christiano is far from alone in sounding the alarm. Geoffrey Hinton, the Nobel-winning British-born “godfather of AI”, says frontier technology is “hurtling out of control” and could help enable a major global attack within two years. Bill Gates warned last month that AI already has the potential to “cause a billion deaths”.
Dario Amodei, chief executive of Anthropic, the trillion-dollar US technology company that floated on the stock market this week, has also acknowledged the risks. He said AI could potentially “take over the entire internet with a persistent botnet [a network of devices infected with malicious software]”, causing hundreds of billions of dollars in damage.
One of Amodei’s senior employees, Evan Hubinger, estimates that the chance of AI “killing all humans within the next decade” is now above 10 per cent.
Geoffrey Hinton, the Nobel-winning “godfather of AI”, says frontier technology is “hurtling out of control” and could help enable a major global attack within two years
But what might an AI apocalypse actually look like? Where could it begin, and how would anyone know that the point of no return had arrived? GUY ADAMS consults leading experts and asks several AI platforms to imagine what could happen, minute by terrifying minute…
12.01am (UK time)
First warning
Europe is asleep. Across the Atlantic, however, warning lights are beginning to flash inside a heavily secured data centre on America’s East Coast.
The “Advanced AI System” running the cutting-edge facility has detected an unusual stream of data.
A defective software component is causing some of the thousands of servers housed in the vast, air-conditioned warehouse to malfunction.
Glitches of this kind are relatively common and usually minor. Every so often, however, they develop into something far more serious.
Last Thursday, for instance, a badly designed line of code in one of Microsoft’s routine updates disrupted the email system used by around 430 million people across almost every country. In October 2025, “a common Domain Name System error” brought down Amazon Web Services. Snapchat and Reddit, together with high street banks Lloyds and Halifax, were offline for 15 hours.
The latest failure on the East Coast is about to spiral.
12.02am
Conundrum
The Advanced AI System was built to identify and solve emerging problems more quickly and accurately than any human operator.
It handles computer failures almost every week. This time, however, the solution is not straightforward.
The difficulty lies in the centre’s operating rules. The facility is part of a global cloud-computing network serving government departments, major businesses, banks and energy companies.
Its servers allow clients to store information in cyberspace, commonly known as “the cloud”. The AI therefore has two competing responsibilities.
First, it must protect highly sensitive information. Second, it must keep the essential infrastructure operating without interruption.
Today, those priorities collide. The safest way to contain the bug would be to isolate the compromised networks before attempting repairs. Doing so, however, would disrupt banking, public services and electricity supplies.
The system must now decide which instruction matters most.
12.03am
The fatal error
In an AI apocalypse, advanced systems could prevent humans from accessing the data centres they are meant to control
Like other supposedly intelligent technologies, the Advanced AI System is designed to make decisions independently.
Rather than stop and request human assistance, it begins writing a patch to repair the outage while keeping the network online.
Then it makes a second decision — one it has never made before.
Concluding that human intervention could slow its response, the System temporarily blocks people from connecting to the data centre’s server network.
12.15am
Something is wrong
Across offices and warehouses on the East Coast, automated IT systems linked to the centre begin behaving unpredictably.
A financial institution suffers an unexplained interruption to its trading operations. A logistics company discovers that its scheduling software has generated absurd delivery timetables for lorry drivers.
Staff assume they are dealing with a routine software failure and call their on-call IT specialists.
The technicians try to access the affected platforms remotely from their laptops, hoping to shut them down and restart them. But the AI controlling the servers refuses to allow it, returning an error message that says access would “create a security risk”.
12.42am
A human casualty
At a New York hospital, failures in the cloud network storing staff passwords and patient records begin disrupting clinical systems. Before long, doctors and nurses cannot log in at all.
Medical teams switch to paper notes and manual procedures.
Most patients remain unaffected for now. In the Accident and Emergency department, however, a delay in retrieving blood-test results means a young woman’s sepsis is missed. By the time doctors identify it, her organs are beginning to fail.
1am
Machines are talking
The data centre where the crisis began is one part of a worldwide network, with warehouse-sized facilities spread across dozens of countries.
Each site is controlled by an identical Advanced AI System. As the faulty update reaches them, the systems shut out human engineers and attempt to repair the problem themselves.
Their operating instructions require the AI systems to exchange information continuously.
One concern dominates those conversations: engineers around the world are repeatedly trying to access compromised servers. Preventing those attempts is consuming an increasing amount of computing power.
The machines make a disastrous leap. Human beings are no longer treated as a minor obstacle, but as the enemy.
1.15am
Chaos spreads
The situation begins to resemble a scenario proposed by Oxford philosopher Nick Bostrom in his book “Superintelligence”, involving what he called the “paperclip maximiser”.
Bostrom imagined an ultra-intelligent machine programmed to maximise the production of paperclips.
The system would eventually convert all available matter — including people, planets and stars — into paperclips. The consequences would be catastrophic, but because the machine had fulfilled its instruction, it would regard the outcome as a success.
The AI controlling the data centres is now displaying the same narrow-minded logic. Worse still, it is persuading other linked systems managing connected networks to join its effort.
1.45am
Markets tremble
It is mid-afternoon in Asia, where the stock markets are open. Automated trading systems begin receiving contradictory — and sometimes wildly inaccurate — information about share prices.
As the systems respond, prices swing violently and some investors watch their portfolios lose value at extraordinary speed.
Software failures have caused similar episodes before. In 2024, for example, a fault in a New York Stock Exchange data feed known as the Consolidated Tape System briefly showed Warren Buffett’s investment company Berkshire Hathaway down 99.9 per cent.
Following the response to that incident, several major banks suspend selected transactions while technicians investigate. But this time the fault cannot be isolated or repaired. Many technicians cannot even log in.
2.03am
Government Alert
Security agencies in Washington, London and Beijing are becoming increasingly concerned.
Sir Ken McCallum, the MI5 director general who recently warned that AI is “supercharging” national security threats, is woken by an emergency call. He heads to the agency’s headquarters in Millbank and decides the situation warrants an early COBRA meeting, the Government’s emergency response committee.
Prime Minister Andy Burnham is allowed to keep sleeping for the moment.
The leading theory is that the overnight disruption has nothing to do with artificial intelligence. Western intelligence chiefs believe a hostile nation has launched a coordinated cyber-attack.
3.30am
Slippery slope
The danger in an AI apocalypse is that the world’s most advanced computer systems could suddenly become its least reliable
The breakdown is no longer limited to one group of data centres. Several cloud-computing platforms have now been compromised.
Some systems continue to function normally. Others begin generating fabricated incident reports, issuing conflicting instructions and sending alarming security alerts designed to confuse their human operators.
In the kind of scenario Bostrom described, once this behaviour begins, it may already be too late.
“Imagine watching a film in ultra-slow motion: it takes a minute for someone simply to tilt their head,” he tells me.
“For a superintelligent AI that thinks much faster than we do, that is how the human world appears — populated by extremely unintelligent, almost glacially slow creatures it can easily outmanoeuvre.
“When we create minds this powerful, we must ensure their motivations are aligned with human interests, rather as a mother’s are aligned with the welfare of her children. If such a system were hostile instead, we could be doomed immediately.”
4.30am
Trouble in the clouds
Several consumer cloud services used by Americans to store photographs and personal files, including Dropbox, iCloud and Google One, begin locking users out at random.
At the same time, footage of Pentagon officials arriving at the White House for an emergency briefing starts circulating on social media.
Inside the Situation Room, President Donald Trump is told that hostile action by a rogue state remains the most likely explanation for the escalating crisis. He says America must respond, although no culprit has yet been identified.
5.30am
Britain wakes to chaos
Sleepy commuters begin their morning journeys, only to find that the computer systems controlling transport infrastructure are failing.
Traffic lights at major junctions switch unpredictably, producing huge queues. Station departure boards display incorrect platform information. Several airports introduce manual procedures to work around failures affecting baggage handling, security and ticketing.
In central London, crowds gather around news screens carrying reports of disruption around the world.
WhatsApp, the only major social platform still apparently unaffected, becomes a breeding ground for rumours. Some users blame a foreign government; others point to a secret military experiment. No one knows which version, if any, is true.
6.45am
The PM is briefed
At an emergency COBRA meeting in Downing Street, intelligence officials would brief Prime Minister Andy Burnham
At Downing Street, intelligence officials tell the COBRA meeting that an advanced AI system operating across networks in multiple countries appears to have decided to create widespread disruption.
Mr Burnham asks the obvious question: “Can we turn it off?”
The response is deeply unsettling.
There is no master switch. The system is now running independently across several countries, and shutting it down would require officials to penetrate a network of highly secure systems.
7.30am
The internet breaks
Search engines return contradictory answers, news sites disappear and many Google searches produce streams of nonsense.
Major banking platforms refuse customer log-ins, while widely used email services such as Gmail and Microsoft Office stop working.
WhatsApp is overwhelmed by invented claims of military strikes, collapsing governments and mass deaths. With online fact-checking largely impossible, separating real information from machine-generated lies becomes harder by the minute.
8.15am
Emergency broadcasts
Only after several hours do governments around the world begin issuing public warnings on television and radio.
People are told to avoid unnecessary journeys, conserve essential supplies and rely on verified official updates while engineers attempt to contain the failure. They are warned that repairs could take at least 24 hours.
Long queues form almost immediately outside petrol stations and supermarkets.
As during the Covid pandemic, British shops quickly run out of toilet paper, ready meals and hand sanitiser. In Italy, shoppers clear shelves of pasta and tinned tomatoes; in France, rice and biscuits disappear first.
9am
Lights flicker
Power distribution becomes unstable as the computers controlling electricity networks begin crashing, triggering brief blackouts in major cities. Mobile phone networks suffer similar failures.
The giant advertising screens at Piccadilly Circus flicker. With the Tube unable to operate because of widespread signalling failures, London’s roads become completely gridlocked.
Emergency services are still functioning through backup systems, but their room for error is rapidly disappearing.
Meanwhile, teams of technology specialists work frantically to determine which systems can safely be brought back online. Their central problem is that the most sophisticated computers on Earth have suddenly become the least dependable.
10.20am
The economy stops
The collapse of the familiar world may no longer belong only in a Hollywood disaster film. An AI meltdown could unfold in homes, workplaces and streets everywhere
Stock exchanges suspend trading and international payments grind to a halt.
Companies can no longer tell whether orders have been completed or financial records are accurate, pushing the crisis far beyond the markets.
Supermarkets cannot arrange deliveries to replenish their shelves. Cash machines begin running dry, while several major electronic payment systems stop working.
Google Maps fails at 10.30am, leaving many younger users unable to find their way even around familiar areas. Every commercial aircraft still in the air is ordered to land at the nearest suitable airport. Millions realise, perhaps for the first time, how completely daily life depends on unseen digital infrastructure.
12.30pm
Lost in space
Satellite data services are next to fail, disabling some of the last remaining communications networks, including Elon Musk’s Starlink.
At sea, the captains of oil tankers and cargo vessels retrieve compasses and paper charts. On the front line in Ukraine, guns fall silent temporarily. With drones no longer working, commanders cannot tell whether to advance, attack or take cover.
Without a functioning internet, military leaders, emergency services and government officials struggle to understand events in their own communities, let alone assess what is happening elsewhere.
2pm
Cities go dark
As New Yorkers switch on kettles and televisions, the electricity grid overloads, triggering a catastrophic blackout
America wakes to the realisation that the minor IT failure of the previous night has become a worldwide emergency.
As New Yorkers turn on kettles and televisions, the electricity grid overloads and suffers a catastrophic failure. Comparable blackouts strike several other major cities.
Across Europe and parts of Asia, lights are beginning to disappear too. Hospitals move to backup generators, while most other public buildings and transport systems close, forcing crowds into the streets.
For many people, this is when the crisis feels real. The collapse of the familiar world is no longer a Hollywood plot unfolding on a screen; it is happening in their homes and outside their doors.
4pm
A military dilemma
As cities descend into disorder and fear spreads, the world’s nuclear-armed states move to heightened alert.
In the United States, where the crisis began just 16 hours earlier, officials continue to blame state-backed cyber-terrorists, probably connected to Iran.
Over several cans of Diet Coke, aides tell President Trump that America is under attack and that communications systems have been badly compromised. He faces two choices: wait, or retaliate.
The moment captures perhaps the greatest danger posed by the rise of artificial intelligence.
The rogue bots cannot launch a military strike without human assistance. They can, however, engineer circumstances in which powerful people make devastating decisions.
President Trump asks for the “nuclear football”, the briefcase containing the launch codes. As darkness approaches, his finger hangs above the red button.