AI Insiders Now Say There's a 1-in-5 Chance AI Ends Humanity
Nearly 1 in 5 AI researchers say AI could end humanity. Should you worry?
Something has quietly flipped in the conversation about artificial intelligence. For years, people who worried that AI could go catastrophically wrong — or even wipe out humanity — were treated as fringe. Now those warnings are coming from inside the room. A long-running survey of AI researchers found the average expert puts about an 18% chance on AI causing human extinction or an equally permanent loss of human control. That's roughly 1 in 5, closer to a coin flip than anyone would like.
The public has moved too. A Politico poll taken over just a few days found about two-thirds of Americans now believe there's at least a moderate risk AI could destroy humanity. At an invitation-only gathering of top US executives at Yale, 93% said they disagreed with the President's claim that AI's catastrophic risks are a hoax. Elon Musk, meanwhile, said he now expects some kind of standalone AI regulator — an agency like the FAA or FCC, but for AI. A DeepMind researcher publicly quit and raised the alarm.
Researchers have a name for this kind of sudden shift: a "preference cascade." It's what happens when people change their stated views because they see others changing theirs, setting off an avalanche. The writer Zvi Mowshowitz argues this cascade has started but isn't nearly enough. Skeptics point out that doom predictions are contested, extinction percentages from surveys are educated guesses, and the loudest opponents of slowing AI down include giant companies like Nvidia with enormous money on the line. Strong opinion, big budgets.
The practical fight is no longer abstract. It's about whether AI development gets slowed, inspected, or regulated — and who pays for that. Supporters of more caution are pushing for action inside AI labs, in media, and in politics. A hastily organized conference on the topic was held in the Bay Area. For everyone else, the takeaway is simple: the people building this technology are less certain it's safe than they used to pretend.
- The average AI researcher now puts about an 18% chance — nearly 1 in 5 — on AI causing human extinction or permanent loss of control
- About two-thirds of Americans say AI poses at least a moderate risk to humanity, according to a recent Politico poll, and 93% of executives at a Yale event rejected the idea that these fears are a hoax
- Elon Musk now backs creating a dedicated AI regulator like the FAA, while major AI companies are spending heavily to fight new rules
Why It Matters
The rules being written now will shape AI's risks and benefits — and you get no vote on the details.