This week, a post on X unleashed panic when Jacob Coxon, announcing his resignation from tech company Anthropic, wrote: “The people building AI earnestly believe that it could kill us all by the end of the decade.” That might have been dismissed as fearmongering had it not been quoted by a research lead at the same company, Evan Hubinger, who replied “Jacob is correct here – we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade.” Alarm ensued.
With some relief, Hubinger subsequently added: “To be clear, as we say in our latest Risk Report, I think the risk from present models is low. What I am worried about is superintelligence.”
These posts come at a time when we have seen a spate of AI agents – software programmes that carry out tasks independently when directed by humans – break out of their test environments and hack into other websites. That sounds concerning, but it’s also attributable to gaps in security and carelessness on the part of the companies testing them. Should we be worried about these particular claims of a threat to human existence? Fortunately, it might be worth taking the doom-mongering with a pinch of salt.
