Researchers are arguing online over the probability that AI could drive humanity to extinction, Axios reported on Tuesday.

The report said a stir erupted this week after it became known that AI researchers had been calculating and debating human-extinction risk in numerical terms. The probabilities presented ranged from 10 percent and 20 percent to, at times, even higher.

The dispute was triggered by a resignation statement posted by Jacob Koxon (제이콥 콕슨), a researcher who worked at OpenAI and then at Anthropic. His post, which argued that frontier AI companies are "gambling with our lives," has surpassed 110 million views on X (Twitter).

The fallout grew after Evan Hubinger (에반 허빙거), who leads Anthropic's Alignment Science team, lent weight to Koxon's claims. He said he genuinely believes AI could kill all of humanity, and he sees the probability of human extinction within the next 10 years as above 10 percent. He added that Anthropic still has no way to keep a superintelligence under human control.

Some interpret it as a publicity mistake, while others see it as fear marketing by Anthropic, which is aiming for a $2 trillion corporate value ahead of an initial public offering (IPO).

The idea of AI-driven human extinction has been raised for several years. Nobel laureate Geoffrey Hinton (제프리 힌턴) estimated the probability of extinction due to AI at 10 to 20 percent. Elon Musk put it at 20 percent. Dario Amodei (다리오 아모데이), Anthropic's chief executive officer, said in an interview with Axios last year that the probability things would go "really very badly" was 25 percent.

Concerning points can be summed up as AI lowering the barriers to designing biological weapons, carrying out cyberattacks or crippling infrastructure, and autonomous systems being able to deceive monitors, self-replicate and resist attempts to shut them down, the report said. The most extreme scenario is AI improving itself faster than humans can control it, Axios reported.

The probability researchers talk about differs from what the public feels. In a lab, 10 percent is closer to meaning "we do not yet really know," but in everyday life people would not accept even that level of risk for an airplane, a new drug or a nuclear reactor, Axios reported.