A top safety researcher at one of the world’s leading artificial intelligence companies just told the public there is better than a one-in-ten chance his own industry’s technology wipes out humanity within ten years.
Quick Take
- Anthropic’s Evan Hubinger publicly said he believes there is more than a 10% chance AI could kill all humans within the next decade.
- Hubinger made the claim while responding to a colleague, Jacob Coxon, who had just resigned from Anthropic over similar fears.
- Hubinger admitted Anthropic has no finished plan to keep a superintelligent AI aligned with human interests.
- Other researchers, including teams at RAND and the Forecasting Research Institute, put the odds far lower, sometimes near zero.
What Hubinger Actually Said
Evan Hubinger works as Anthropic’s Alignment Science Lead, the team tasked with making sure powerful AI systems stay under human control.
He posted on X that “we really do earnestly believe AI could kill all humans” and put his own estimate at over 10% within the next decade. He added that Anthropic is “trying its best,” but admitted the company has “no plan to solve alignment for superintelligence” yet.
That is not a fringe outsider talking. Hubinger sits inside the company building some of the most advanced AI models on Earth.
His comments landed hard because he is not guessing from the sidelines. He is describing the honest internal odds, as he sees them, from someone paid to prevent the very disaster he is describing.
The Resignation That Sparked The Debate
Hubinger’s post came right after a former colleague, Jacob Coxon, resigned from Anthropic and accused AI labs of racing toward self-improving systems without adequate safeguards.
Coxon reportedly said companies including Anthropic and OpenAI are “gambling with our lives”. Hubinger’s response was meant to validate, not dismiss, that warning, which is notable given his job is literally to reassure the public that Anthropic has safety under control.
More than 10% chance AI 'could kill all humans' – Anthropic researcher https://t.co/TuJwPLi2CB
— Sky News (@SkyNews) September 9, 2026
This is not a story about a lone doomsayer. It is two insiders, one departing and one still on payroll, agreeing that the risk is real and unresolved.
For an industry racing to build ever-more-capable systems while competing for billions in investment, that kind of internal alarm deserves serious attention rather than a shrug.
Why Other Experts Think The Odds Are Much Lower
Researchers at RAND studied the same question and reached a very different conclusion. One senior RAND scientist called human extinction from AI “possible but highly unlikely,” and the team wrote that no scenario exists where AI is “conclusively an extinction threat to humanity.”
Their reasoning is simple: humans are spread across the whole planet and adapt fast, making a total wipeout mechanically hard to pull off with any known technology.
The Forecasting Research Institute found an even bigger split. Skeptical forecasters put the odds of AI-caused catastrophe by the year 2100 at roughly 0.12%, while researchers who already worried about AI risk estimated 20% to 25%.
That is not a small disagreement, that is two camps looking at the same technology and landing worlds apart.
A separate poll of AI engineers found a median extinction estimate of just 5%, and a University of Melbourne roundup found three out of five experts said AI poses no existential risk at all, though it noted Nobel laureate Geoffrey Hinton still estimates a 10% to 20% chance over the next thirty years.
This actually happened…This morning’s AI headline is almost impossible to process: an Anthropic researcher just resigned saying the people building AI genuinely believe it could “kill us all by the end of the decade.” Anthropic’s alignment lead says he puts the risk above 10%.…
— Edward Corona – The Options Oracle (@EdwardCoronaUSA) September 9, 2026
What should a common-sense reader take from this mess of numbers? Nobody, including the people building these systems, actually knows the real odds.
When insiders at a top AI lab put double-digit extinction risk on the table while outside researchers argue the number rounds to zero, that gap alone is the story.
It argues for caution, transparency, and real congressional oversight, not blind trust in any single company’s internal safety promises, no matter how well-intentioned those promises sound.
Sources:
cbsnews.com, seekingalpha.com, en.wikipedia.org, static1.squarespace.com, consumeraffairs.com
