Anthropic researcher warns: Artificial intelligence could wipe out humanity, risk exceeds 10% within a decade

Evan Hubinger, who leads artificial intelligence alignment and safety research at Anthropic, estimates that the likelihood of AI causing the extinction of humanity over the next ten years is higher than 10%.

“We sincerely believe that artificial intelligence could kill everyone. Personally, I think the likelihood is above 10% over the next ten years,” Hubinger wrote on the X platform. He stressed that although Anthropic is trying to address the issue, the company still has no solution for aligning a superintelligence with human interests and values. According to him, it is not even clear whether they are moving closer to such a solution.

The position became public after researcher Jacob Coxon left his job. Coxon had previously worked at OpenAI and Anthropic. He criticized both companies, accusing them of engaging in a race to create a superintelligence capable of improving itself and of “playing with our lives.”

However, Hubinger distinguished between current systems and those that may emerge in the future. He said the risk posed by the models used today is low, while the main concern is linked to systems that could self-improve and escape human control.

The estimate of more than 10% reflects Hubinger’s personal opinion and is not an official forecast by Anthropic, Forbes and The Wall Street Journal wrote while commenting on the artificial intelligence researcher’s startling statement.