WASHINGTON: Artificial intelligence researcher Jacob Coxon has resigned from Anthropic, warning that leading AI companies are “gambling with our lives” as they race to develop increasingly powerful systems.
Coxon, who previously worked at OpenAI, said in a post on X that the power of AI should not be underestimated, warning that future systems could become capable of hacking computer systems, transforming entire fields and acquiring significant real-world resources and influence.
He said Anthropic and OpenAI were “racing straight to self-improving superintelligence”, referring to AI systems capable of developing more capable successors and potentially accelerating their own progress.
“The people building AI earnestly believe that it could kill us all by the end of the decade,” Coxon said in a follow-up post.
Evan Hubinger, an Anthropic staff lead working on AI alignment, backed Coxon’s assessment, saying the company does believe AI could potentially pose an existential threat to humanity.
Hubinger estimated the probability of such an outcome at more than 10 percent over the next decade and said there was currently no clear plan for ensuring AI remains aligned with human objectives in a superintelligence scenario.
The warnings come amid growing international debate over the risks posed by increasingly autonomous AI systems. The European Union’s AI regulations require companies to assess and mitigate risks associated with loss of human control over advanced AI systems.
Coxon also called for a moratorium on advancing AI capabilities, arguing that the industry needs stronger safeguards before developing more powerful systems.
The concerns contrast with the rapid commercial expansion of AI companies. Anthropic has attracted major investments and is reportedly considering an initial public offering, although the company has not publicly confirmed specific valuation targets or a listing date.
Meanwhile, more than 1,100 AI professionals, including Anthropic Chief Executive Dario Amodei, have backed the “Pacing the Frontier” initiative, which calls for mechanisms to slow AI development if technological progress begins to outpace regulatory oversight.