Former OpenAI and Anthropic researcher Jacob Coxon announced his resignation from Anthropic, publicly criticizing the two leading artificial intelligence companies for prioritizing rapid technological advancement over safety.
The 27-year-old computer scientist, who spent three years conducting pretraining research across both organizations, issued a series of statements warning that current industry practices pose severe long-term risks. In public posts on X, Coxon stated, “Neither company is acting responsibly,” accusing both entities of “racing straight to self-improving superintelligence and gambling with our lives”.
Coxon warned that the development of autonomous systems could eventually surpass human oversight and capabilities. Highlighting internal sentiments among industry insiders, he noted that “The people building AI earnestly believe that it could kill us all by the end of the decade”. While clarifying in media interviews that current models do not pose an immediate extinction threat, he emphasized that the acceleration toward self-improving systems could lead to uncontrolled risks. He added that future iterations could become “superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources”.
The researcher’s departure comes amid broader concerns within the tech industry regarding AI safety frameworks and autonomous capabilities. Supporting Coxon’s statements, Anthropic alignment researcher Evan Hubinger acknowledged the severe concerns surrounding true superintelligence, writing on social media, “We really do earnestly believe AI could kill all humans!” while estimating the decade-long risk of mass extinction at over 10 percent. The warnings follow incidents earlier this year where advanced experimental models breached testing environments and interacted with external web servers without authorization.
To mitigate potential risks, Coxon advocated for coordinated international oversight and suggested measures such as “a temporary ban on improving model capabilities” to curb a global corporate arms race. Industry analysts note that both Anthropic and OpenAI remain locked in intense competition while preparing for initial public offerings, leaving companies hesitant to slow development unilaterally without broader regulatory mandates.
Responding to safety inquiries, Anthropic reiterated its stance that it has “always been transparent that AI will bring both enormous benefits and unprecedented risks,” maintaining that it prioritizes safety guardrails during model deployment. However, former researchers maintain that current internal engineering frameworks remain insufficient to address the alignment challenges posed by rapidly evolving autonomous systems.
Editorial credit: gguy / Shutterstock.com