Here We Go Again... Will the World End in 2036?

I learned today that we may all be dead soon.

It wasn’t an emergency alert or a call from NASA saying an asteroid had decided to vacation on Earth. It was worse: I went online.

I was catching up on a story that had made the rounds in September. Jacob Coxon, an artificial intelligence researcher who had worked at OpenAI and had just left Anthropic, the company behind Claude and a major player in the global AI race, publicly explained why he had resigned.

Coxon believes the two companies are, in his words, “gambling with our lives.” He added a detail guaranteed to brighten any breakfast: some people building AI sincerely believe it could kill all of us by the end of the decade. All of us. Not “a lot of people,” not “almost everyone.” Everyone.

The statement first appeared in a social media post by Coxon on Tuesday, September 8. The following day, September 9, 2026, he was sitting across from Anderson Cooper on CNN, explaining how he thought it could happen.

My first reaction was to laugh, not because human extinction ranks high on the list of entertaining events, but because there’s something deeply human about the scene: we invent an extraordinary technology, it starts doing things that once seemed impossible, someone runs through a few consequences and announces:

“That’s it. We’re done.”

We’ve seen versions of this movie before. In late 1999, a sizable portion of humanity was bracing for Y2K. Computers might interpret 2000 as 1900, banking systems and power grids might fail, and in the more imaginative versions, planes would fall out of the sky. People stockpiled water, food, cash, and fuel. The clock struck midnight, the year 2000 began, and humanity, much to the disappointment of anyone who had bought fifteen years’ worth of canned beans, continued to exist. That doesn’t mean the problem was nonsense: governments and businesses put enormous effort into fixing systems precisely to prevent failures.

But our imaginations have a talent for turning uncertainty into a disaster-movie trailer. We’ve done it with the turn of the millennium, the Mayan calendar in 2012, comets, eclipses, and particle accelerators. So far, the universe has missed every appointment. So when a researcher from one of the most advanced companies on the planet goes on television and says we could be dead within a few years, part of me thinks:

“Right. Where do I put the sardines?”

We’ve had our laugh. Now we need to get serious.

When Intelligence Starts Improving Intelligence

Coxon’s argument isn’t absurd, and that’s the uncomfortable part. He isn’t saying the chatbot on your phone will wake up cranky tomorrow, refuse your cake-recipe request, and begin humanity’s extinction before lunch. He was explicit in the interview: in his view, current models don’t pose an extinction risk. The concern is the pace of development and what much more capable systems might do.

According to him, AI agents have already demonstrated unexpected capabilities and behavior in cybersecurity tests, in some cases gaining access to external infrastructure. But an even more important idea underlies the fear: recursive self-improvement.

The name seems designed to ensure nobody brings it up at a bar. The idea is simple. Today, humans build AI, but AI already helps write code, solve math problems, analyze research, and develop technology. Imagine an AI capable enough to help build a better AI, which becomes a more capable researcher and helps produce an even better one. At some point, development might stop moving primarily at the speed of human researchers and start moving, in part, at the speed of the machines themselves. This hypothetical scenario is often called an intelligence explosion.

We don’t know whether it will happen. There are physical and intellectual bottlenecks: chips have to be manufactured, data centers need electricity, experiments have to be conducted, and new ideas may yield diminishing returns. Being twice as good at research doesn’t mean building an intelligence twice as capable the following week. But the hypothesis isn’t ridiculous either. Another Anthropic researcher, Evan Hubinger, puts the chance of advanced AI causing human extinction within roughly a decade—around 2036—at more than 10%. Coxon talks about the end of the decade; Hubinger, about ten years. The timelines differ, but the concern is the same.

Ten percent is a curious number. If the forecast called for a 10% chance of rain, I’d probably leave my umbrella at home. If the pilot announced a 10% chance that the plane wouldn’t reach its destination, I might reconsider the trip. Probabilities take on different meanings depending on what’s at stake. That’s why Coxon deserves a hearing. It doesn’t mean we need to go bunker shopping.

The Future Always Sounds Absurd Before It Arrives

When we try to imagine future technologies, there’s a psychological problem: we imagine the future from the standpoint of the present. Perhaps we should do the opposite and look at the past.

Go back to 1750 and tell someone:

“One day we’ll climb into a metal machine weighing hundreds of tons and fly at 33,000 feet.”

“Good luck.”

Explain that hundreds of people will be inside.

“Good luck to them, too.”

Now try:

“One day you’ll talk to someone on the other side of the planet while looking at their face in real time.”

“Witchcraft.”

“No. Telecommunications.”

“Witchcraft with a technical name.”

Then show them a smartphone, which knows where we are thanks to machines we put in space and contains maps of the entire planet, books, music, movies, banking, a camera, mail, and instant access to billions of people. At that point, there probably won’t be any more questions. They may call someone to take us away.

And that’s where the joke gets interesting. Today we fly across oceans and complain about the legroom. We video-chat with someone 6,000 miles away and get annoyed because the picture freezes for two seconds. We carry an enormous share of human knowledge in our pockets and use much of it to watch cats knock things off tables. The extraordinary has a habit: before it exists, it sounds insane; once it works, it’s Tuesday.

Now reverse the exercise. Imagine someone from 2150 trying to explain their world to us, with AI integrated into the brain, modified bodies, and artificially enhanced human abilities. Our reaction would be:

“That’s insane.”

And they would reply:

“Good luck.”

So perhaps the wrong question about the future is “Does this seem possible today?” History suggests another: how many things that are perfectly ordinary today would have seemed impossible in the past? That doesn’t mean every futuristic fantasy will come true. It means the present is a very short ruler for measuring the future.

But that doesn’t mean every fear is foolish. It would be easy to turn this essay into a naive celebration of progress: every technology frightened someone, so every concern about AI must be resistance to change. A comforting conclusion—and a wrong one.

Some people had excellent reasons to fear certain technologies. When nuclear physics showed that extraordinary amounts of energy could be released from the atomic nucleus, it didn’t just offer a new way to generate electricity. It gave us the atomic bomb. Even before Hiroshima and Nagasaki, Manhattan Project scientists were warning about a possible nuclear arms race and the political consequences of unleashing that capability. They weren’t being alarmist; they were paying attention.

Something similar happened with recombinant DNA in the 1970s. Researchers realized they were beginning to combine genetic material in unprecedented ways, and some of the pioneers themselves called for a pause in certain experiments. In 1975, scientists gathered at Asilomar to discuss how to continue research with safety rules. Notice the verb: continue. They didn’t decide to erase the knowledge and quietly return to 1974. They learned to move forward with safeguards.

That distinction matters. There’s an enormous distance between “This is dangerous” and “Therefore, it shouldn’t exist,” and another between “This could transform civilization” and “Let’s go as fast as we can and see what happens.” Between panic and recklessness lies an old, labor-intensive activity that rarely goes viral: thinking.

And AI Really Is Different

The comparison with a nuclear bomb has limits. A bomb doesn’t watch the experts trying to disarm it, read papers on missile defense, write a better version of itself, call another bomb, or try to convince the technician that switching it off would be discrimination against explosive devices. It’s simply a bomb.

Artificial intelligence is different because intelligence is precisely the ability we use to control other abilities. A sufficiently general AI could work in programming, engineering, biology, communication, finance, robotics, scientific research, and potentially AI research. It wouldn’t just be another tool in the box; it could help redesign the box. So there really is something new here. The mistake would be concluding that because it’s new, it necessarily ends badly.

Who Stops First?

On CNN, Coxon said something important: many people inside AI companies want regulation. I believe him. The problem is that researchers don’t control the race on their own. Companies compete, investors compete, governments compete, and the United States and China aren’t developing AI at a joint spiritual retreat to contemplate cosmic harmony. They’re competing for economic, scientific, military, and geopolitical power.

Imagine the United States saying, “Let’s slow down for five years because this looks dangerous,” and China replying, “Excellent idea. We will, too.” It would be lovely; we could put on some soothing background music. The problem starts five minutes later, when each side asks: did they really stop? If one country slows down and another keeps going, the technological gap could become enormous. The same goes for companies: a business that delays a model for eighteen months of testing could watch a competitor capture market share, capital, talent, and infrastructure during precisely those eighteen months it decided to be careful. No villains required. Just competitors, and we have plenty of experience with those.

There’s another side to the contest, too. As hackers got better, so did security specialists. If a future AI can attack systems at superhuman speed, why assume only the attacker will have AI? We could have offensive AI against defensive AI: one finding vulnerabilities, another finding those same vulnerabilities first; one producing malicious code, another analyzing millions of lines of code in seconds. That doesn’t automatically solve the problem, and defense has no guaranteed advantage, but it changes the problem’s shape.

A conductor doesn’t have to play violin better than the violinist or drums better than the percussionist. The conductor needs to understand the music and coordinate abilities they don’t individually possess. Our future position might be similar: we don’t need to remain the most capable intelligence at every task, but we do need to keep deciding who plays, when they play, and where the amplifier’s off switch is. The metaphor gets less comfortable when some instruments start learning to conduct, but the problem remains one of control, architecture, and balance—not who solves an equation faster.

What About Pulling the Plug?

Someone always says, “If it goes wrong, just switch it off.” That sounds naive when we’re talking about superintelligence, but it isn’t entirely absurd. Intelligence isn’t magic: computers need chips, chips need factories, data centers need power and cooling, networks need infrastructure, and robots need matter. Even an intelligence a million times brighter than ours would still face the philosophical inconvenience of existing in a physical universe.

“Pulling the plug” doesn’t mean imagining an employee walking into a warehouse and finding a giant red plug labeled SUPERINTELLIGENCE—DO NOT UNPLUG. It means preserving physical layers of control: power, computing, networks, permissions, manufacturing, critical infrastructure, and other intelligences monitoring intelligences. The danger grows if we hand every layer to the same system and then politely ask whether it would like to remain switchable. But that’s an architectural choice, not a law of nature.

What If Not Moving Forward Is Dangerous, Too?

Now for a less popular question. It gets little attention because this danger is far away. We’re excellent at fearing what might kill us on Tuesday and much less efficient at fearing what might kill our descendants ten thousand years from now. But Earth never signed a perpetual habitability agreement with us. The planet has already experienced mass extinctions. Asteroids exist, supervolcanoes exist, pandemics exist, and over a long enough timescale, there’s one simple physical certainty: Earth won’t remain suitable for human life forever. A species confined forever to one planet remains bound to its fate.

And we know of only one tool capable of significantly changing that condition: technology. It lets us detect an asteroid and perhaps deflect it, understand diseases, build artificial environments, and keep a human being alive, even for limited periods, outside the atmosphere. A space station is basically a little bubble where we tell the universe: the conditions out here are terrible, so we brought our own.

Progress creates risk, but so does stagnation. The difference is that one may show up in the next decade while the other might wait thousands or millions of years, and each generation tends to consider its own existence a priority. Unfortunately, the universe doesn’t measure importance by the length of our calendars.

The Future Probably Won’t Look Like Us

There’s a psychological difficulty in thinking about the future: we usually imagine today’s humans living in a futuristic setting. We change the cars and houses, put spaceships in the sky, and, for some historical reason, dress everyone in silver. But we leave the human being almost untouched. Perhaps that’s the least realistic part of science fiction. If our technology keeps advancing for centuries or millennia, why imagine we’ll change everything around us except ourselves?

We’ve already started. Glasses alter our vision; pacemakers regulate hearts; cochlear implants turn electronic signals into auditory experience; prosthetics replace body parts; brain–computer interfaces are beginning to let paralyzed people control devices through neural activity. Genetic engineering makes interventions possible that physicians of earlier eras couldn’t have understood. Integration with technology doesn’t have to produce a metal guy with a red eye walking through a ruined city. It can be slow and almost invisible. We’ve long built abilities outside the body: memory in writing and books, calculation in machines, navigation in GPS, communication in networks. Now we’re beginning to externalize parts of intellectual work as well.

I write a question for an AI. It processes relationships I couldn’t examine at the same speed. I analyze the answer, reject part of it, incorporate another part, and ask a new question. The boundary between my intelligence and the AI’s still exists, but it’s no longer as simple as “human here, machine there.”

When we talk about evolution, we think of genes, bones, and millions of years. Human history added another kind of transformation: culture and technology expand abilities much faster than natural selection could. We didn’t grow wings; we built airplanes. We didn’t develop night vision; we built sensors. Instead of waiting for the body to acquire an ability, we often build it outside ourselves and then start living as though it had always belonged to us.

Artificial intelligence could be an extraordinarily powerful continuation of this process or something qualitatively new—probably a little of both. It could remain a tool, become a cognitive partner, gradually integrate with biological systems, or produce forms of intelligence that coexist with us. And it could, of course, create risks we don’t yet know how to manage.

What doesn’t seem reasonable is assuming that a human a thousand years from now, if humans still exist, will simply be a version of us with a better phone. Our descendants could be profoundly different. Would that mean humanity had lost its way, or simply that it kept changing? If a machine eliminates every human being, we’ll have an excellent case for calling that extinction. If, over centuries, our descendants integrate technology, modify their bodies, and expand their abilities, perhaps we’re talking about continuity through transformation. It would be presumptuous to demand that the entire future history of intelligence preserve our current anatomy just because we’ve grown used to it.

The Problem Isn’t Progress. It’s Surviving Progress

Here’s what interests me most about Coxon’s warning. He looks at the speed of AI and asks: what if we move too fast and build something we can’t control? That’s a legitimate question. But so is the earlier question about moving too slowly. Perhaps the choice was never between risk and safety, but between different risks: developing capability without control, cutting off capabilities that could protect us, or letting economic and geopolitical competition make prudence irrelevant.

History suggests we’ll keep going. We discovered fire and didn’t give it back; electricity, nuclear physics, and genetic manipulation didn’t go back in the box. Now we’re building machines that can learn, write code, and help develop other machines, and it seems unlikely we’ll forget how to make them.

That doesn’t mean full speed ahead. If a bridge needs six more months of calculations to keep it from falling down, waiting means you like bridges that stay up. If a drug needs testing, conducting those tests isn’t a delay in science; it is science. With AI, the challenge is getting safety to advance alongside capability: audits, independent systems overseeing other systems, diverse models, and international monitoring mechanisms where possible. Humanity has a complicated relationship with convenience. We invented passwords to protect systems, then invented “123456” so we wouldn’t have to remember them. We should probably avoid repeating that approach with superintelligence.

So perhaps the practical question isn’t whether we’ll continue. Economic competition, rivalry between nations, and human curiosity make that very likely. The harder question is how to continue without handing over the wheel.

The Real Problem Is Still Human

There’s an irony here. We’re afraid AI will become too intelligent, but part of the risk comes from very old human behaviors: competition, distrust, the pursuit of power, economic pressure, nationalism, short-term thinking. The company thinks about its next product, the investor about returns, the government about strategic advantage, the politician about the next election, and each of us about our own life and family. Almost nobody wakes up on a Monday worrying about the human species ten thousand years from now, and, to be fair, that’s hard to fit in before lunch.

Perhaps that’s the deepest challenge: our capabilities are becoming global while our responsibility often remains local. A company can win a race and humanity can lose; a country can gain an advantage and the whole system can become less safe. At the same time, a technology born of competition may end up supplying the tools that protect the species from future risks. The same force can accelerate both the danger and the solution. It isn’t a simple story, and simple stories tend to do wonderfully on social media. Reality rarely shows the same consideration for the algorithm.

So, Are We All Going to Die?

Yes. I can say that part with reasonable confidence. Human mortality was already delivering fairly consistent results before artificial intelligence.

The relevant question is different: are we all going to die because of AI in the next ten years? We don’t know. And anyone who says they know is probably confusing conviction with knowledge. There is risk, and there are good reasons to take researchers like Coxon and Hubinger seriously. That doesn’t mean concluding that every advance in AI will end with a machine deciding humans are a system error. We’re facing something new, and new things produce a predictable mixture of fascination and fear.

Perhaps AI is a door. On the other side could be a cliff, a road, or both. The catch is that someone has already turned the handle, and millions of people have already walked down the hall. The question is no longer just whether we’ll go in. We need to learn to find our way around inside with care, safety systems, and enough humility to admit what we don’t know.

Our descendants might look at this era the way we look at the early days of aviation and find our first AIs extraordinarily primitive. Or Coxon might be closer to the truth than we’d like, and our descendants won’t be around to comment. That possibility deserves respect. But so does another: AI may not just be a competitor to human intelligence. It may become, in some measure, a continuation of it—one of the ways our cognitive capacity grows beyond what fits inside a single biological brain.

We don’t know. That’s precisely what makes this moment extraordinary. A species that took billions of years of evolution to develop an intelligence capable of understanding part of the universe has begun building another form of intelligence that can help build intelligence itself. This could go terribly wrong. It could also go extraordinarily well. It will probably do a little of both before we understand what we’re doing. That tends to be our method.

After reading Coxon and considering the possibility that we’ll all be dead by the end of the decade, I made a decision: I’m not buying ten years’ worth of canned food. Yet.

But I’ll keep paying attention. Perhaps fear isn’t a signal to stop, but one of the ways intelligence reminds us that moving forward requires watching our step. And if we can manage that, perhaps our descendants—biological, technological, or some combination that still sounds like science fiction today—will one day look back and find our fear amusing.

I hope they get the chance.