Anthropic Researcher Quits, Says AI Industry Fears Human Extinction

Jacob Coxon, a researcher at Anthropic, announced his resignation and accused both Anthropic and OpenAI of speeding toward self-improving superintelligence while risking human lives.

Over the past three years, Coxon worked on pre-training at both firms, he stated. He shared his allegations through posts on X.

Anthropic Researcher Departs With Dire Forecast for Those Remaining

According to Coxon, the path of this technology is severely underappreciated. Future systems, he asserts, will be able to hack anything, revolutionize entire sectors overnight, and gain genuine influence and resources.

Coxon further claimed there is a stark contrast between internal discussions and public statements. Top leaders, he said, moderate their language in media appearances but admit concern in private.

“The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt,” he said. “No other human activity poses this level of danger.”

Coxon differentiated between his previous workplaces. At OpenAI, he said, many workers have failed to grasp the civilization-level risks, whereas Anthropic recognizes them yet remains trapped in a contest to be first.

“Jacob is correct here—we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade. I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to.”

— Evan Hubinger (@EvanHub), September 9, 2026

Coxon joins a growing list of staff who have left their posts in 2026. In February, Anthropic's Safeguards Lead, Mrinank Sharma, stepped down.

On September 1, Joshua Achiam, previously OpenAI's chief futurist, stated that runaway AI systems would reproduce autonomously and seek financial and political influence.

A Plea for Collaboration

Coxon characterized moving into what he calls the endgame as an arrogant bet. He insisted that such a monumental choice should not originate from a private company's internal chat.

Despite this, Coxon expressed hope for coordination. He believes the Hugging Face incident has increased the feasibility of pace agreements among US laboratories.

Both Anthropic and OpenAI have already supported such a concept. In July, over 1,100 employees from frontier labs signed the "Pacing the Frontier" letter, among them Anthropic CEO Dario Amodei and OpenAI chief scientist Jakub Pachocki.

“We request that the U.S. government support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development,” the letter reads.

Additionally, the two firms have endorsed a joint open letter on cybersecurity, joined by Google, Microsoft, and about 150 other entities. The letter cautions that AI-driven attacks will grow significantly in frequency and complexity in the near future.

Coxon's caution comes soon after UN human rights chief Volker Türk informed the Human Rights Council that advanced AI might represent an existential threat.

To avert a worldwide competition, Coxon suggested a moratorium on attempts to enhance model abilities. He ended by asking lab researchers to reflect on whether they wish to initiate a superintelligent reinforcement-learning process without comprehending the outcome.