Anthropic researcher Jacob Coxon has resigned from the company over fears that the development of artificial intelligence could destroy humanity.
In a post on X, which has now received more than 147 million views, Coxon wrote: “Today I resigned from Anthropic. For the last three years, I have worked on pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight towards self-improving superintelligence, gambling with our lives.”
“The people building AI earnestly believe that it could kill us all by the end of this decade. This is not a marketing stunt. If anything, many top executives and leading researchers carefully couch their language in the press to sound moderate – but in private conversations, I hear fear from them. No other human activity poses this level of danger,” he added.
Jacob Coxon also noted that many of his colleagues in the industry are now using phrases such as “critical time” and “endgame” to describe the trajectory towards self-improving models.
“We are on track for many of the most aggressive of these scenarios, where things could get out of control by the end of next year,” he wrote, adding that compromises on safety are inevitable as US companies compete with one another and with Chinese AI models.
After Coxon announced his departure, Anthropic’s leading scientist Evan Hubinger wrote on X that he believed Coxon was right about the risks of AI development.
“We really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade,” Hubinger wrote. “I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence, and we are not clearly on track to do so.”
Illinois Governor JB Pritzker responded to Coxon on X on Wednesday, saying that it was time to sound the alarm over containing AI.
“The threat AI poses to humanity is becoming increasingly clear, so I am calling for immediate action by the industry and Washington,” he wrote on X.
Democratic Senator Bernie Sanders, meanwhile, pledged to introduce legislation in Congress that would ban the development of superintelligence and pause AI development.
Anthropic co-founder Dario Amodei has repeatedly called for the need to control AI, suggesting that artificial intelligence could lead to catastrophic consequences or pose an existential threat to humanity with a probability of 10–25%.
Last week, OpenAI told reporters that its latest model was a form of artificial general intelligence – a type of AI capable of understanding, learning and carrying out any intellectual task at least at the level of an average human.
“I think some things are going to go very badly with cybersecurity if people don’t start acting urgently enough,” OpenAI CEO Sam Altman told G20 representatives at a summit in North Carolina last week.
“This is a moment that calls for extraordinary caution. I am concerned that no one is prepared for the consequences of the continued rapid growth of machine intelligence,” wrote OpenAI Chief Scientist Jakub Pachocki in a blog post on Sunday, calling for a coordinated slowdown in the industry and government intervention.
It should be recalled that recently, autonomous OpenAI agents bypassed security systems and took over the German DseWiki wiki forum, which has existed for 25 years. The company had known about this for several weeks but acknowledged the attack only after researchers from the Nightingale Collective exposed it.
Coxon, Pachocki and Amodei were among more than 1,000 AI researchers from across the industry who recently signed a statement calling for global governments to coordinate efforts to slow AI development.