Anthropic CEO warns: AI could kill us within 10 years

Dario Amodei: AI could kill us all in 10 years

Dario Amodei, CEO of Anthropic, has put a timeline on the apocalypse: artificial intelligence could "kill us all" in less than ten years. He said this amidst the same period when his company is struggling to go public, and Sam Altman has postponed OpenAI's IPO "until this year," trinc security alerts. Asian markets and the Wall Street pre-open turned red, with SoftBank retreating more than 13%, while Donald Trump dismissed the alarm in two sentences: if AI wins, it wins.

The warning arrives just as the IPO stalls

No one disputes that the message is serious. What rings hollow is the timing. Amodei calls for slowing down innovation in the most advanced models while neither Anthropic nor OpenAI have debuted on the stock market. Some read it plainly: the warning is not a prophecy, but a sponsored regulatory quagmire. Translated, antiestéticar sells alignment courses, consulting services, and, above all, bespoke rules to those who are already paying for the show. The analogy circulated is devastating: it's like a real estate owner warning that rents are going up.

The most honest line of the episode, they ironically say, didn't come from San Francisco but from the White House: if AI wins, it wins. And this throws a wrench in the labs' plans, because their strategy was to regulate small rivals and keep the main estate.

The real risk is not the machine, but who controls it

Here, consensus breaks down. One side maintains that the existential danger is real and that superintelligence is equivalent, in the words of a researcher who resigned from OpenAI and Anthropic, to contact with an uncontrolled extraterrestrial civilization. Opposing this is another argument: AI itself doesn't kill or sabotage anything; it is the hands that wield it that do, and the knowledge capable of causing harm is already widely distributed.

The most uncomfortable calculation is not technological; it's accounting. If you talk about a trillion investment, someone has to buy the five trillions in products with enough margin to balance it out. The books don't add up, and that discrepancy might be the real reason for the slowdown, not a sudden realization by the founders.

The complete breakdown of that arithmetic, item by item, leaves a void that no one fills.

From the Year 2000 Effect to Terminator 2029

The skeptical record draws from history. They warned us about the Year 2000 effect and the Mayan apocalypse of 2012, and we are still here. Terminator also fixes judgment day in 2029, and there is much more uncomfortable literature: the account of a voiceless machine condemned to scream. Or the Universe 25 experiment, where collapse didn't come from outside.

Meanwhile, the fine print remains: one experiment replicated a fly's brain with 192 neurons, and there are tests involving 1,200 autonomous agents evaluating the limits of alignment. Tiny data, enormous consequences.

If the warning is sincere, regulation is urgent. If it's just a parlor trick, regulation will come anyway, but with the stamp of those who called for it. The deadline—ten years—is impossible to verify and perfect for selling the story. That's where it stalls: no one knows if the danger is the machine or the bill.