Anthropic Researcher Resigns, Warns AI Could Kill Humans

Researcher exits AI safety firm over existential concerns

Jacob Coxon, an artificial intelligence researcher who worked at both Anthropic and OpenAI, has resigned from Anthropic with a stark warning: the leading AI companies are "gambling with our lives."

In a public post announcing his departure, Coxon cautioned that AI systems will soon become superhuman in capability, able to "hack anything, revolutionize any field overnight, and acquire real power and resources." He urged observers not to underestimate the technology's trajectory, noting that progress shows no signs of slowing.

According to Coxon, both Anthropic and OpenAI are racing toward self-improving superintelligence—a scenario where AI models can autonomously develop more capable versions of themselves, creating an unstoppable feedback loop. This concept is viewed by researchers at companies like Google-owned DeepMind as a potential trigger for artificial superintelligence, where AI vastly exceeds rather than merely matches human intelligence.

Why it matters

Coxon's resignation is particularly significant because Anthropic was founded explicitly as a safety-focused alternative to OpenAI. When even researchers at companies positioning themselves as responsible AI developers express existential concerns publicly, it signals a widening gap between the pace of AI development and the maturity of safety frameworks. The fact that current staff members at Anthropic are confirming rather than disputing these warnings suggests the concerns are not fringe views but mainstream assessments within the field.

Internal confirmation of existential risk

In a notable development, Evan Hubinger, Anthropic's staff lead on AI alignment—the effort to keep AI systems working toward human goals—publicly validated Coxon's assessment. "Jacob is correct here — we really do earnestly believe AI could kill all humans," Hubinger wrote, though he has not left the company.

Hubinger estimated the probability of this outcome at greater than ten percent within the next decade. He acknowledged that no plan currently exists for maintaining AI alignment in a superintelligence scenario.

Coxon emphasized in follow-up remarks that those building these systems "earnestly believe that it could kill us all by the end of the decade."

Legislative response emerging

The warnings come as policymakers begin responding to superintelligence concerns. Last week, U.S. Senator Bernie Sanders announced plans to introduce legislation that would ban companies from developing superintelligence.

These details were first reported by Politico.