Anthropic Researcher Quits After 4 Months, Warns AI Could Kill Us All By Decade’s End

Anthropic researcher Jacob Coxon has resigned just four months after joining the AI company, walking away before his equity vested and warning that the race toward self-improving artificial intelligence could pose catastrophic risks to humanity.

Coxon, who previously worked at OpenAI, accused both companies of racing toward increasingly powerful AI systems without sufficient guarantees that humans will remain in control. Axios reported Coxon left before vesting.

 

TL;DR

 

• Jacob Coxon quit Anthropic after only four months and two months before his company equity was due to vest.
• He warned that developers believe advanced AI could threaten humanity before the decade ends.
• Anthropic researcher Evan Hubinger put the extinction risk above 10% within a decade.
• Coxon wants AI labs and governments to coordinate on slowing self-improving AI development.

 

Why Did Jacob Coxon Quit Anthropic?

 

Coxon announced his resignation after spending the previous three years conducting pretraining research across OpenAI and Anthropic. TechCrunch covered his resignation over safety.

He had been at Anthropic for only four months. According to Coxon, employees must remain for six months before their stock begins vesting, meaning he left two months before receiving any Anthropic equity.

“I left before any of my equity vested,” Coxon told Axios.

His resignation comes with a much broader warning about the direction of the AI industry. Coxon claimed that Anthropic and OpenAI are racing toward self-improving superintelligence while lacking reliable methods to guarantee that highly capable systems remain aligned with human intentions.

He also said he had not personally seen Anthropic compromise its safety practices. Instead, his concern is that increasingly fierce competition could eventually force frontier AI companies to trade safety rigor for development speed.

“If you're under pressure to race, you have to cut corners,” he said.

 

Coxon Warns AI Could Threaten Humanity

 

Coxon's concerns extend beyond existing AI models to future systems capable of improving themselves and operating at superhuman levels across areas such as software development, cybersecurity and scientific research.

In an interview, WIRED examined Coxon’s warnings about superintelligence. Coxon described the next one or two years as “crunch time for humanity,” arguing that decisions made by frontier AI labs could determine whether increasingly powerful systems can be safely controlled.

He pointed to possible AI-enabled cyberattacks and biological threats as examples of how sufficiently advanced systems could create catastrophic consequences if alignment fails.

Coxon's warnings are not isolated inside Anthropic.

Evan Hubinger, who leads alignment science at the company, publicly backed the broader concern and estimated that there is a greater than 10% probability that advanced AI could cause human extinction within the next decade. Financial Times reported Anthropic researchers’ concerns.

Hubinger stressed that the danger from today's models remains low, with his concerns focused primarily on future systems capable of recursively improving their own abilities.

What Does Anthropic Say?

 

Anthropic has consistently positioned AI safety as a central part of its strategy and acknowledged that increasingly capable artificial intelligence could create unprecedented risks.

Responding to Coxon's concerns, an Anthropic spokesperson told WIRED that the company believes the industry would benefit from a lawful and verifiable framework for coordinating how powerful AI models are released.

Coxon himself also described Anthropic as considerably more serious about AI safety than OpenAI based on his experience at both companies.

However, he argued that responsible behavior by one company may not be enough if competitive pressure forces multiple AI labs into a race where slowing development means risking losing technological leadership.

 

Could AI Development Be Slowed?

 

Coxon has called for coordination between Anthropic and OpenAI to prevent an immediate race toward recursive self-improvement, where AI systems help create increasingly capable successors.

He believes broader international cooperation would eventually be required, particularly between major technological powers such as the United States and China.

For now, his resignation adds another insider voice to a growing debate surrounding whether frontier AI development can move quickly enough to remain competitive while moving slowly enough to keep increasingly autonomous systems under meaningful human control.