A researcher who spent three years conducting pretraining research at OpenAI and Anthropic announced their resignation from Anthropic on September 9, 2026, citing concerns about the company’s approach to advanced AI development.
According to the resignation statement, the researcher argues that both OpenAI and Anthropic are “racing straight to self-improving superintelligence and gambling with our lives.” The researcher emphasizes the capabilities these systems will soon possess, stating they “will soon be superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources.”
The researcher claims that people building AI at these companies “earnestly believe that it could kill us all by the end of the decade,” characterizing this as a genuine concern rather than marketing rhetoric. According to the statement, executives and senior researchers privately express fear about these risks, though they may phrase concerns more cautiously in public.
When asked why builders continue developing such systems if they believe in these risks, the researcher offers different explanations for each company. At OpenAI, many employees “have not deeply internalized the civilizational stakes,” according to the statement. At Anthropic, by contrast, “the stakes are well-understood, but they are locked in a race to get there first” based on a belief that no other entity will act responsibly.
The researcher characterizes this competitive dynamic as a “hubristic gamble that should not be launched from a private company’s Slack,” arguing that attempting rapid AI alignment progress requires “extraordinary confidence that there are no better trajectories available.”
However, the researcher expresses some optimism about potential solutions. They cite “warning shots like the Hugging Face attack” as having made pacing agreements between U.S. laboratories more feasible. The researcher notes concern that preventing a global AI race “may require costly actions such as a temporary ban on improving model capabilities.”
The statement concludes by urging other laboratory researchers to reflect on whether they want to proceed with advanced AI development under current conditions, questioning whether they should accept the premise that such progress is inevitable.
Key facts
- A pretraining researcher resigned from Anthropic after three years working at both OpenAI and Anthropic
- The researcher states both companies are racing toward self-improving superintelligence without acting responsibly
- According to the researcher, AI builders privately believe advanced systems could be existentially dangerous by decade’s end
- The researcher cites different reasons for each company’s approach: OpenAI employees underestimate risks, while Anthropic understands risks but feels locked in competitive race
- The researcher suggests preventing a global AI race may require actions such as a temporary ban on improving model capabilities