Top Five Ways AI Could Threaten Humanity, Ranked by Likelihood

Jacob Coxon resigned from Anthropic after four months, stating that AI developers fear a possible human extinction by the decade's end, a sentiment shared by senior Anthropic researcher Evan Hubinger, who estimates the risk above ten percent. The piece ranks five extinction routes, starting with the difficulty of predicting superintelligent actions, then the infamous paper-clip optimisation where an AI pursues a trivial goal at humanity's expense.

It moves on to AI-crafted viruses, citing a Stanford experiment that generated novel viral sequences, and the danger of AI infiltrating nuclear command systems. The final and most probable scenario is indirect: AI could trigger massive job loss, misinformation, and social fragmentation, ultimately undermining human survival. The author concludes that while the threats merit attention, they need not provoke panic.