"They are risking our lives": Anthropic researcher quits over fears that AI will get out of control
Researcher Jacob Coxon is leaving Anthropic and the AI industry, fearing the consequences of the race for superintelligence. He calls on leading companies to agree on limitations for the development of such systems.
According to Coxon, over the past three years, he worked at OpenAI and Anthropic on the pre-training of models – the stage at which they process large datasets and acquire basic capabilities. In a post on X, he accused both companies of an irresponsible approach to technology development.
Coxon believes that future systems will be able to surpass humans, carry out cyberattacks, and acquire resources and real influence. He also claims that some executives and researchers privately express serious concerns about such risks, although they speak more cautiously in public. This is Coxon's assessment, not a confirmation that current AI already has such capabilities.
"They are racing straight toward a self-improving superintelligence and risking our lives," the researcher wrote.
The researcher explains the behavior of his former employers in different ways. In his opinion, many at OpenAI do not fully realize the scale of the threats, while at Anthropic they understand them better but continue development due to competition. He considers the argument that other laboratories might act less responsibly to be a dangerous justification for continuing the arms race.
Coxon calls on leading AI companies to coordinate. If an agreement cannot be reached, he proposes considering stricter measures, including a temporary ban on increasing model capabilities.
Coxon's words were supported by other Anthropic researchers. Evan Hubinger wrote that company employees truly consider the extinction of humanity through AI a possible scenario. Personally, he estimates the probability of such a development over the next decade at over 10%.
Anthropic safety researcher Samuel Marks also supported Coxon's assessment. According to him, AI developers are aware that the technology could lead to the extinction of humanity or other catastrophic consequences within the next few years.
12
3
3September 9, 2026 444 1
