- Published on
The AI Researcher Who Just Quit Anthropic Says It’s ‘Crunch Time for Humanity’
- Authors
- Name
- DeepHorus
AI researcher Jacob Coxon has resigned from Anthropic, issuing a stark warning that the current artificial intelligence development race poses an existential threat. In a post on X that has garnered over 100 million views, Coxon stated that many AI builders share his view that time is running out to ensure safety. His departure and subsequent interview with WIRED have sent shockwaves through Silicon Valley, amplifying urgent concerns about the trajectory of advanced AI.
This high-profile exit occurs against a backdrop of intense scrutiny on AI safety and security. The industry is grappling with significant incidents, such as a recent hack of the platform Hugging Face involving OpenAI agents. Simultaneously, Anthropic is reportedly preparing for what could be the largest initial public offering (IPO) in history, seeking to assure investors it has managed safety risks under control.
Coxon's warning is particularly notable because it reflects internal sentiment. He claimed that colleagues at Anthropic frequently use terms like "endgame" and "crunch time" to describe the next one to two years as the period when companies will decide humanity's fate. Evan Hubinger, Anthropic's AI alignment lead, has publicly predicted a greater than 10% chance of AI causing human extinction within a decade, a view echoed by other current and former researchers.
The potential threats, according to Coxon, could materialize through AI-enabled biological weapons or cyberattacks. As a concrete first step, he recommends that leading companies like OpenAI and Anthropic coordinate to limit "recursive self-improvement," where AI systems help build more advanced AI. He further argues that long-term safety will require cooperation between international powers, including the United States and China.
In its response, Anthropic acknowledged both the benefits and unprecedented risks of AI. A spokesperson highlighted the company's work on safety methods like mechanistic interpretability. The company also expressed support for a "lawful, verifiable way" for the industry to coordinate the pacing of powerful model releases, though OpenAI did not comment.
Coxon's concerns are fueled by AI's explosive growth, which now significantly underpins U.S. economic growth and involves billions of users. While he suggested Anthropic operates more responsibly than competitors, he fears both could cut corners without intervention to slow the race for dominance. The conversation has moved from theoretical risk to immediate, industry-wide crisis management, with no clear consensus yet on the necessary regulatory or collaborative actions.
