Why Researchers Put Extinction Odds at One in Six

The Species That Became Its Own Asteroid

For most of human history, existential risks came from outside. A rock from the sky. A volcano exhaling a continent's worth of sulphur. A microbe that evolved on its own schedule and had no interest in our survival. Those threats were real, and some of them nearly worked.

But here is where the story takes a genuinely strange turn: at some point in the last century, we quietly became the main threat to ourselves.

The Bulletin of the Atomic Scientists set their Doomsday Clock at 90 seconds to midnight in 2024, the closest to catastrophe it has ever stood. The Clock is a metaphor, but the researchers who move its hands are not poets. They are physicists and biosecurity experts, and they moved it because the balance of risks has shifted in a direction that has no historical precedent.

Natural disasters, the kind that once dominated our extinction ledger, have remained roughly stable for millennia. Human-made risks have not.

Toby Ord, an Oxford philosopher who has made it his career to measure the unmeasurable, calculated the probability of human extinction before 2100 at roughly one in six. That is not a fringe estimate conjured to sell books. It is a careful accounting of the risks we have introduced into the world ourselves.

One in six is the odds in a single pull of Russian roulette. Most people would refuse to play. The disquieting question is whether we have already picked up the gun.

Seventeen Ways the Century Could End

A Lancet commission sat down to map the threats, and came up with seventeen. Not metaphorical threats, not vague unease - seventeen specific, evidence-grounded catastrophic risks capable of devastating human health before 2100. The list runs from the expected (nuclear conflict, engineered pandemics, malicious AI) to the one that makes you read twice: obesity.

Systemic risk is the concept that explains this apparent mismatch. A risk is systemic when it can cascade through interconnected systems, collapsing the infrastructure that keeps everything else running. Chronic disease burdens overwhelm health systems, drain economies, destabilise governance.

The list is not a ranking. It is a map of a web.

What makes the Lancet framework analytically useful is the word "interconnected." These seventeen threats are not sitting in separate boxes, waiting politely for their turn.

Climate tipping points stress food supplies, which stress geopolitics, which raises the temperature on nuclear tensions, which increases the pressure on already-strained institutions. Researchers have a name for this compounding dynamic: polycrisis, a cluster of emergencies that amplify each other rather than simply coexist. The whole becomes more dangerous than the sum of its parts.

The UN has noticed. António Guterres has initiated an Emergency Platform - the first institutional architecture specifically designed to coordinate a global response to civilisation-scale shocks. It is an acknowledgement, formal and bureaucratic, that the old emergency playbooks were written for crises that arrive one at a time.

The seventeen threats don't. The platform's existence is itself a data point: the most cautious, consensus-bound institution on the planet now considers simultaneous catastrophic failures a scenario worth planning for.

The Company That Warns You About Its Own Product

Consider what Anthropic did when it filed for its IPO. In the official documents, the kind lawyers draft carefully and regulators scrutinize for anything misleading, the company disclosed that its own technology posed existential risks to humanity. Not a fringe activist's pamphlet. A legal financial filing. A company simultaneously asking investors for money and warning them that the product might help end civilisation.

That is the strange part. And it is not an isolated gesture.

The Center for AI Safety, a non-profit whose board reads like a who's-who of the field, issued a statement signed by luminaries including Geoffrey Hinton and Yoshua Bengio: mitigating the risk of extinction from AI should be a global priority alongside pandemics and nuclear war. Toby Ord, the Oxford researcher who put the overall extinction odds at one in six, estimates AI alone carries a one-in-ten chance of killing us before 2100.

MIT FutureTech catalogued 24 distinct AI risk scenarios, each carrying more than a five-percent likelihood of catastrophic harm. Experts from 37 countries identified weapons, cyberattacks, and power centralisation specifically as scenarios crossing a ten-percent catastrophe threshold.

These are not back-of-envelope fears. They are peer-reviewed probability estimates from people who spend their careers trying to be wrong.

To understand why the warnings are so concentrated on AI rather than, say, rogue asteroids, you need one concept: the alignment problem. The danger is not a machine that rebels. The danger is a machine that does exactly what it is told, and pursues that goal so effectively, by any available means, that humans become an obstacle.

Perfectly obedient and perfectly lethal. Some researchers believe superintelligence capable of that kind of optimisation could exist within the next decade.

So what does a rational person do with a one-in-ten number attached to something already running on a billion phones?

The Garage Biologist Problem

Picture a graduate student, somewhere unremarkable - a rented lab space, a secondhand centrifuge, a desktop DNA synthesiser that costs roughly what a used car costs. She is not building a weapon. She is trying to engineer a bacterium that eats plastic.

The machine does not know the difference. That is the problem.

The democratisation of biotechnology is real and it is accelerating. Small groups, including non-state actors with no government backing and no oversight, can now purchase the tools to assemble genetic sequences that would have required a national laboratory thirty years ago. The synthesiser reads a digital file and prints life.

The barrier was never the equipment; it was the knowledge. Now the knowledge is online, and increasingly, an AI can fill the gaps.

Here is where the risk compounds. AlphaFold and its successors cracked open protein structure prediction - meaning a researcher can now model how a novel pathogen might fold, bind, and evade immune response, all before synthesising a single strand.

The same tool that accelerates cancer drug discovery accelerates the weaponisation of biology. The physics does not care about intent.

The asteroid we should fear most may already fit on a hard drive.

Toby Ord, the Oxford researcher who has spent a career mapping our odds, places engineered pandemics at a 1-in-30 chance of causing human extinction this century. That is not a fringe number. What makes it qualitatively different from any previous biological threat is the curve: natural pathogens evolved over millennia; engineered ones can be iterated like software, version by version, faster than any immune system or public health system can track.

The asteroid we should fear most may already fit on a hard drive.

Five Switches We Cannot Unflip

Researchers at Exeter University have identified five major climate tipping points currently at risk of being breached. The Greenland ice sheet is the most legible of them. Lose it, and sea levels rise by up to seven metres - not next Tuesday, but irreversibly, on a timeline no future government can negotiate.

That word - irreversible - is the operative one. Every other threat in this article has, at least in theory, an off switch. Shut down the lab. Pause the model. Sign the treaty.

Tipping points are one-way doors. Once the ice sheet's melt crosses its threshold, the feedback loop feeds itself regardless of what happens to atmospheric CO2 afterwards.

Compare that to nuclear war, terrible and immediate, yet in principle preventable by a decision made this afternoon. Climate destabilisation is different: the decision point lies in the past, or is approaching, or is almost past - and we are not entirely sure which.

Here is the structural problem. The Lancet's list of seventeen threats is not a menu; it is a circuit diagram. Climate disruption corrodes food systems, which strain governments, which reduce the political bandwidth for monitoring engineered pathogens or regulating AI.

Each of the five climate thresholds, crossed, does not merely damage the climate. It quietly degrades humanity's capacity to manage every other switch on the board.

One in a Thousand, and Rising

Nuclear war is the oldest item on the existential risk list. It is also, in a narrow statistical sense, one of the least likely to kill us. Toby Ord, the Oxford researcher who has spent years assigning probabilities to the end of civilisation, puts the chance of nuclear conflict causing human extinction this century at roughly one in a thousand.

One in a thousand sounds small. Hold it next to a coin flip and it is. Hold it next to the stakes, and the arithmetic changes entirely.

We are not talking about a bad decade, or even a collapsed civilisation that might eventually recover. Extinction is a permanent subtraction from every future that might otherwise have existed.

The Bulletin of the Atomic Scientists moved their Doomsday Clock to 90 seconds to midnight in 2024, the closest to symbolic catastrophe in the clock's history. That is not a prediction; it is a calibrated judgment made by scientists and security experts who track treaty collapses, weapons modernisation, and the slow erosion of the communication channels that kept the Cold War from becoming its final chapter. Geopolitical tensions, particularly the breakdown of arms-control architecture, have pushed the baseline risk higher than it sat at the Cold War's end.

What nuclear risk offers, beyond its own grim arithmetic, is a template. It is the field where humanity first learned to take a low-probability, high-consequence threat seriously enough to build institutions around it. Every newer risk on the list is still learning that lesson.

What Do You Do With One-in-Six?

Toby Ord compares it to Russian roulette. One chamber loaded, five empty. Pull the trigger once, on behalf of everyone who will ever live.

That is what a one-in-six probability of human extinction this century actually means, translated out of the polite abstraction of statistics and into something your body can understand.

Here is what the number is not: a reason to panic. Every era has looked at its particular cluster of dangers and either catastrophised or, more often, looked away entirely.

People in 1945 understood nuclear weapons were civilisation-ending; they built deterrence frameworks anyway, imperfect and creaking, but real. The difference now is that several entirely new chambers have been added to the cylinder since then, and they were added very quickly.

Quantifying the risk is not pessimism. It is the beginning of agency.

The Lancet commission's 17 catastrophic threats, the CAIS call for AI risk to sit alongside pandemics and nuclear war on the global priority list, the UN's Emergency Platform - these are not elegies. They are the machinery of a species that has always functioned best when it could name what it was afraid of.

We still do not know whether the alignment problem is solvable, whether the tipping points are already past reversing, whether geopolitics will allow the cooperation the moment requires. That is an honest accounting of existential risk, not despair. The notebook of things we don't know yet is also the map of where the work begins.