Jacob Kaxon, a prominent artificial intelligence researcher, took an unusually cryptic step upon leaving Anthropic; unlike many of his peers who depart to launch new startups or protest promotional business models, he saw his resignation as an opportunity to issue a public warning about the existential risks posed by leading AI structures. In a social media post, Kaxon claimed that large AI companies are effectively gambling with the lives of billions by developing systems that possess “genuine beliefs” and could become the factor killing all humans by the end of this decade.
According to him, this danger does not lie so much in current models as in the awareness of the imminent emergence of “Self-improving Superintelligence.” These superintelligent systems will have the capabilities to hack any system, transform virtually any field overnight, and gain access to real power and resources. Kaxon believes that others are either developing these models without comprehending the existential importance of this issue, or they believe that progress toward Superintelligence should be accelerated to prevent critics or irresponsible actors from reaching it first.
These criticisms are not limited to a single individual; Evan Hubinger, the head of the Alignment Science division at Anthropic, endorsed this viewpoint on social media. Hubinger wrote: “Jacob is right here — we really do believe that AI could wipe out all of humanity! I personally estimate the likelihood of this disaster occurring in the coming decade to be greater than 10%.” These statements highlight deep fissures and serious concerns within the leading technical layers of the AI industry.