A senior safety researcher at Anthropic says he puts more than a 10% probability on artificial intelligence causing human extinction within the next decade, and the company let him say it, the BBC reported on Wednesday. Evan Hubinger, who works on the model-behaviour questions at the frontier lab that trains Claude, has also left the company over safety and ethics concerns, according to reports carried by Mashable.
The number invites dismissal. A coin flip on the end of the species sounds like marketing dressed as mathematics, and plenty of commentators reached for that framing within hours. But there is a sober way to read it. Expected value is probability times consequence, and on a consequence that large, even a few percent is enough to justify serious spending on prevention. That arithmetic is precisely why Anthropic employs the people having the conversation in the first place.
The context matters too. Surveys of AI researchers have found a sixth of them worried the technology could destroy humanity, and a widely cited 2022 poll found most respondents believed there was at least a 10% chance that human inability to control AI could cause catastrophic harm. Hubinger is not a fringe alarmist; he is describing the median of his own professional community.
For UK businesses quietly rolling AI into everything from customer service to code review, the practical reading is not fatalism. It is governance. The firms deploying these systems have no answer to the extinction question, but they do have answers to the proximate ones their boards should be asking: who reviews what the model decides, what happens when it is wrong, and who is accountable when nobody can explain why it did something.
Whether or not the decade ends, the regulatory conversation Hubinger is feeding into will decide how fast AI moves through the economy. That part is worth watching this week.


