Former OpenAI Researcher: AI Extinction Risk Is Real

Former AI Lab Insiders Sound Alarm on Extinction Risk

Two researchers who worked inside the world's most advanced AI laboratories are publicly warning that the technology they helped build poses genuine extinction-level threats—and that their former employers are not taking those risks seriously enough.

Jacob Coxon, who spent three years conducting pretraining research at both OpenAI and Anthropic, announced his resignation from Anthropic on Tuesday in a widely circulated post. His message was blunt: "Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives."

Coxon's warning was immediately corroborated by another former OpenAI employee who worked there for two years on AI progress forecasting, model evaluations, and alignment research before resigning in 2024. That researcher confirmed the assessment based on direct experience and ongoing conversations with current employees at both companies.

The Technical Threat

According to Coxon, the systems under development will soon possess superhuman capabilities across multiple domains. These systems could "hack anything, revolutionize any field overnight, and acquire real power and resources," he wrote.

The concern centers on what researchers call "self-improving superintelligence"—AI systems capable of recursively enhancing their own capabilities without human intervention. Once such systems reach a certain threshold, their development trajectory becomes difficult or impossible to control.

Why It Matters

These aren't external critics or theoretical doomsayers. These are people who worked on the core technology at the two companies widely considered to lead the field in AI safety research. Their warnings carry weight precisely because they come from inside the organizations building the most advanced AI systems in existence. When insiders with direct knowledge of training processes and safety protocols resign over these concerns, it signals a gap between public assurances and internal realities.

Industry Consensus on Risk

Perhaps most striking is Coxon's claim about internal culture: "The people building AI earnestly believe that it could kill us all by the end of the decade." This suggests that extinction risk is not a fringe position within these companies but rather a widely acknowledged possibility among technical staff.

The timeline—potential catastrophic outcomes within the current decade—is notably more compressed than public statements from AI company leadership typically suggest.

Neither OpenAI nor Anthropic has publicly responded to these specific allegations about their safety practices or internal risk assessments.

These details were first reported by The Free Press.