Ten Years of Extinction Warnings, and the AI Race Never Slowed
From Stephen Hawking in 2014 to a researcher quitting Anthropic last week, doomsday predictions have become a recurring feature of the AI industry. They haven't stopped it.
Key points
- Stephen Hawking warned in 2014 that AI "could spell the end of the human race", nearly a decade before the public could use ChatGPT.
- Anthropic researcher Jacob Coxon resigned recently, publicly declaring that human extinction from AI was imminent.
- A decade of high-profile warnings from scientists and tech leaders has not visibly slowed investment or development.
- If the warnings don't change the outcome, they may be doing something other than changing the outcome.
In 2014, Stephen Hawking told the BBC that artificial intelligence "could spell the end of the human race." It stopped nobody.
Last week, Jacob Coxon, a researcher at Anthropic, the AI safety company, resigned and went public with the same essential fear: human extinction from AI, and soon. The story, first reported by The Guardian, spread fast. The labs kept building.
We have followed Coxon's resignation across four stories since 9 September, including the moment his warning reached the Senate floor.
Why do the warnings keep coming?
The people making them are often insiders. Hawking was a world-renowned scientist. Coxon worked at a company whose founding mission is AI safety. These aren't fringe voices.
The warnings share a shape. A credible person sees something that frightens them, speaks publicly, and the industry acknowledges the concern, points to its safety work, and carries on. New models ship.
That reflects a genuine disagreement about timelines. Most researchers believe dangerous superintelligent AI, meaning AI far smarter than any human and capable of acting against human interests, is a real risk but decades away, if it arrives at all. Coxon believes it is close enough to justify stopping now.
What does this mean for ordinary people?
Most of the risk that affects people today is not superintelligence. Scams, fake images of real people, targeted phishing emails, automated fraud: that harm is happening now.
The extinction debate pulls public attention toward a dramatic distant threat and away from the concrete present one. A nurse who reads that AI might end humanity in thirty years feels abstract dread. That same nurse, learning that AI tools are already impersonating her hospital to steal patient data, has something she can act on today. Those are very different states of alertness.
A decade of doomsday warnings has been good at generating headlines and poor at changing behaviour, inside the industry or out.
Should you take the next warning seriously?
Yes, selectively. When a senior figure at an AI lab resigns over safety concerns, it is worth attention, not because it confirms catastrophe, but because insiders see things the public doesn't. Watch whether Coxon's specific concerns prompt any response from Anthropic beyond its standard safety statements.
The more revealing signal is the gap between what AI companies say about safety and what they actually build and release. As our earlier reporting found, the numbers researchers put on extinction risk vary wildly and the framing matters enormously. That gap, between stated caution and actual behaviour, is where the real story lives.
