Jacob Coxon, a researcher specializing in training artificial intelligence models, announced his resignation from Anthropic and issued a stern warning about the direction the industry is taking. He explained that both Anthropic and OpenAI are racing towards self-improving artificial intelligence systems, while the safety mechanisms are still not prepared to manage their consequences.
Coxon stated that he spent the last three years conducting pre-training research at both companies. In a series of posts on X, he argued that neither is acting responsibly and accused the industry of “gambling with our lives” in a race to develop superintelligence.
The accusation quickly went viral, reaching tens of millions of views.
“AI could kill us all before the decade is over”
One of the most concerning points in Coxon's statements was his claim that people working on artificial intelligence development seriously believe that the technology could lead to the extinction of humanity before the decade ends.
The researcher maintained that there is a difference between what some professionals express publicly and what they acknowledge privately.
According to his allegations, many executives and researchers soften their statements when speaking with the press but engage in much more alarming internal discussions about the risks associated with developing increasingly powerful systems.
Coxon also warned that upcoming systems could achieve superhuman capabilities in numerous areas, including the ability to breach computer systems, rapidly transform industries, and gain access to resources and power.
The race between OpenAI and Anthropic
Coxon's concern is not limited to one company. The researcher pointed directly at OpenAI and Anthropic, two of the leading artificial intelligence labs in the world.
According to his analysis, the problem is that both companies are caught in a technological race to be first. If one company slows down development for safety reasons, there is fear that another will continue to advance and ultimately gain a decisive advantage.
In the case of Anthropic, Coxon stated that the company understands the risks better but continues to move forward due to competition. Regarding OpenAI, he claimed that some members of the company have not sufficiently internalized the consequences that this race could have.
An Anthropic researcher also supported the warning
Coxon's statements gained even more relevance after Evan Hubinger, head of alignment research at Anthropic and Coxon's former supervisor, publicly backed part of his warnings.
Hubinger stated that the possibility of an AI causing human extinction within the next decade exceeds, in his own estimation, 10%. At the same time, he acknowledged that Anthropic is trying to address the issue, but there is still no sufficiently clear plan to resolve the so-called alignment problem of superintelligence.
Alignment is precisely one of the major challenges of advanced AI: ensuring that systems much more capable than humans continue to act according to the goals and limits set by people.
Coxon calls to halt the race
In light of this scenario, the researcher called for greater coordination among the laboratories developing artificial intelligence and even proposed a temporary ban on increasing the capabilities of the most advanced models.
His argument is that it is not enough to assume that development will continue simply because other companies are also advancing. In his view, the industry should pause to evaluate what could happen when systems reach intelligence levels far superior to the current ones.
The warning comes at a time when OpenAI, Anthropic, and other companies are competing to develop increasingly autonomous and capable models, while researchers and specialists continue to debate the extent to which existing safety mechanisms will be sufficient.
Coxon's case thus exposes a central contradiction of the AI revolution: those building the most powerful systems are also some of those warning that we still do not know how to fully control them. And now one of those researchers has decided to leave the race precisely because he believes the risk has become unacceptable.
