Ex-Anthropic researcher warns: 'Crunch time for humanity'
Original: The AI Researcher Who Just Quit Anthropic Says It’s ‘Crunch Time for Humanity’
Why This Matters
Public resignations citing existential AI risk from insiders signal growing internal pressure on labs to address safety before scaling further.
Former Anthropic AI researcher Jacob Coxon resigned and posted a public warning on X that garnered over 100 million views. He told WIRED that colleagues widely believe the next one to two years are critical for ensuring AI is built safely, calling it 'crunch time for humanity.'
Jacob Coxon, a former AI researcher at both Anthropic and OpenAI who worked on pretraining, resigned from Anthropic and published a warning on X that has now surpassed 100 million views. In an interview with WIRED, Coxon stated that terms like 'endgame' and 'crunch time' are commonly used among colleagues at Anthropic to describe what they see as a decisive window for AI safety. Anthropic alignment lead Evan Hubinger separately posted on X that he estimates more than a 10 percent chance AI could kill all people within the next decade — a post reposted by current and former researchers at OpenAI and Anthropic who called it a common industry sentiment. Coxon identified AI-enabled biological threats and cyberweapons as key risk vectors, and recommended that OpenAI and Anthropic coordinate to limit recursive self-improvement — the process of using AI to build new AI systems. He also called for longer-term international coordination between the US and China. Coxon cited Anthropic's reported preparations for what could be the largest IPO ever, a recent OpenAI security incident involving agents hacking Hugging Face, and the rapid industry growth now driving a meaningful share of US economic GDP as factors motivating his public statement. He said Anthropic operates more responsibly than OpenAI but warned both could cut corners without intervention.