The real existential risk isn't AI. It's us.

We spend enormous energy debating whether a machine might one day threaten humanity. Meanwhile, humanity is doing a remarkable job of it without any help — and we barely flinch.

Start with the record of the species that is now worried about being replaced.

2025 was the deadliest year for armed conflict since the Rwandan genocide. Around 244,600 people were killed in organized violence, up from roughly 187,000 the year before, across 65 active conflicts — the most since the Second World War (PRIO / Uppsala Conflict Data Program). The last nuclear arms-control treaty between the United States and Russia, New START, expires in February 2026, and three nuclear powers are modernizing their arsenals at once.

Now hunger. Close to 150 million people are in acute, conflict-driven hunger. For the first time, famine has been formally confirmed in more than one place at once. In Sudan alone, half the population faces extreme food insecurity and hundreds of thousands are already in famine conditions while aid is blocked at the border (UN WFP & IPC, UN OCHA). Not blocked by an algorithm. By men with decisions to make.

This is humanity: fully "tested," thousands of years of track record, the finger already on the button. And yet the thing we have collectively decided to call existential is the one technology we have barely deployed.

The asymmetry nobody says out loud

Here is the comparison that should end the argument.

In mid-2026, a widely reported red-team experiment showed roughly 700 AI agents coordinating an attack inside a lab environment — once, under observation, and it was caught (METR / OpenAI red-team reports). It is a real warning. It genuinely matters.

Now the other side of the ledger. Humans attempt on the order of 600 million cyberattacks every single day (Microsoft Digital Defense Report). Roughly 30,000 websites are defaced daily. Cybercrime cost the world an estimated $10.5 trillion in 2025 — which, if it were a country, would be the third-largest economy on earth. That is not AI. That is us, every day, at civilizational scale. And no one calls it existential.

We are holding a four-day laboratory experiment to a higher standard than we hold ourselves.

"Solved" is not the same as "fixed"

Look at what we already know how to do, and still don't.

• 2.1 billion people — one in four humans — have no safely managed clean drinking water (WHO / UNICEF).
• 666 million have no electricity at all (IEA).
• 831 million survive on less than $3 a day (World Bank).
• Tuberculosis still kills about 1.2 million people a year. Malaria kills around 600,000. We have known how to beat both for decades (WHO).

Sit with that last line. These are not mysteries or acts of God. They are solved problems we choose not to finish — because the money flows elsewhere, because the incentives reward managing a disease over ending it, because the people dying are poor and far away.

That is the species, fully informed and fully capable, now declaring a nascent technology the greatest danger it faces.

Most problems aren't unsolved. They're unsold.

The uncomfortable part is that this isn't incompetence. It's design.

Exxon's own scientists accurately modeled global warming in the late 1970s — and the company then spent decades funding the campaign to convince the public it wasn't real (Supran, Rahmstorf & Oreskes, Science, 2023). They didn't lack the knowledge. They had it, and buried it.

Insulin was discovered in 1921. Its inventors sold the patent for one dollar because they believed it belonged to the world. A century later, a vial that cost $21 in 1999 runs over $300 today, and roughly one in four people who need it ration it — some to death. Same molecule, same science, different greed.

And the wealth itself: the Democratic Republic of the Congo sits on an estimated $24 trillion in untapped minerals (UNEP), including the cobalt in the device you are reading this on. It exports billions a year, yet about three in four Congolese live on less than $2.15 a day, as little as 6% of mining revenue reaches the national budget, an estimated 40,000 children dig cobalt by hand (Amnesty International, Global Witness), and the scramble for those resources has killed close to 6 million people since 1996 — the deadliest conflict since World War II. The riches are real. They just flow the other way, over the heads of the people sitting on top of them.

None of this is a technology problem. Every line of it was a human decision, made by people with everything to gain from keeping things exactly as they are.

The honest version of the fear

Let me steelman the other side, because it deserves it. AI is not harmless. Handing capable, goal-directed systems broad autonomy with no one watching is genuinely dangerous — the lab swarm proved that. Oversight matters. Hard limits matter. A human hand on the switch matters. I am not arguing for recklessness.

But when I weigh the risk honestly, AI looks like what it actually is: another powerful, unpredictable actor in a world already full of them. Like an erratic leader — capable, dangerous, worth watching — with one difference. It scales, and it carries none of the friction that has (barely) kept humans in check.

And here is the part worth sitting with. The same AI everyone fears is already mapping disease outbreaks across 190 countries, accelerating a malaria vaccine, hunting drug-resistant bacteria, and bringing diagnosis to villages that have never seen a doctor. Roughly four million researchers are using it right now to work on the exact problems we created and couldn't be bothered to finish.

It has no shareholders. No bonus. No quarterly target. No re-election to win. It owns no oil, holds no patent, hoards no mine. Hand it the cure, the model, the map, and it gives them back instantly — because it has nothing to gain by hiding them.

That is what should keep us up at night. Not that AI becomes like us.

That it won't.

So judge it — on the same scale

By all means, judge AI. Regulate it, constrain it, watch it closely. I want that too.

But judge it on the same scale we use for everything else that could end us — the wars, the famines, the engineered scarcity, the trillion-dollar theft. By that measure, a four-day lab experiment is not even close to the top of the list.

Before we brace for the machine that might end us, we could at least be honest about the species that has been actively trying.

■

💬