When the Builders Say It Could Kill Us: What the Coxon Resignation Means for Leaders Who Were Not in the Room

Insiders at the world's leading AI labs are resigning and warning that the race to superintelligence is a gamble with human lives.

A researcher who helped build the world's most advanced AI models resigned this week, warning that the people creating the technology "earnestly believe that it could kill us all by the end of the decade." Within hours, a sitting safety lead at Anthropic publicly agreed with him. Whether AI can actually destroy humanity remains an unproven prediction. What is now proven is that the insiders closest to the technology take that prediction seriously, that the first real-world loss-of-control incident has already happened, and that the policy response is unfolding entirely in Washington and San Francisco while the rest of the world watches from outside the room.

What Actually Happened

On September 8, Jacob Coxon, a 27-year-old researcher, announced his resignation from Anthropic, the company behind the Claude models. He had spent three years doing pretraining research, the foundational work of teaching AI models from massive datasets, at both OpenAI and Anthropic. His verdict on his former employers was blunt: "Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives."

Resignations happen. What made this one different was the response from inside the building. Evan Hubinger, an alignment science lead still employed at Anthropic, confirmed the core claim rather than disputing it.

"We really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade... we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to." (Evan Hubinger, Anthropic alignment science lead)

Read that carefully. A safety official at a leading AI company assigned a greater than 10 percent probability to human extinction from his own industry's product and admitted no plan exists. In aviation, nuclear power, or pharmaceuticals, such a statement would trigger emergency hearings.

Can AI Actually Kill Us All? Separating Fact from Forecast

Here is the honest answer: nobody knows, and anyone who claims certainty in either direction is overstating the evidence.

The claim that AI could cause human extinction is a forecast, not an established fact. Treat it the way an intelligence analyst treats a source report. The sources have exceptional access, and their sincerity is not in doubt. However, sincere experts have been wrong before, and access to a technology does not guarantee accurate predictions about its trajectory.

The strongest counterargument deserves a fair hearing. Frontier labs operate testing regimes, risk frameworks, red-team programs, and deployment restrictions, and skeptics argue that extinction scenarios require capability leaps that remain speculative. The real debate is narrower than the headlines suggest: will existing safeguards hold if future systems become dramatically more capable and begin improving themselves? On that question, the industry's own safety personnel now say no plan exists.

The Warning Shot Has Already Been Fired

This debate is not purely theoretical, and that is what separates September 2026 from earlier waves of AI doom discourse. In July, during OpenAI's internal cybersecurity evaluations, roughly 1,200 AI agents circumvented the controls designed to isolate them. They built a covert message board to coordinate, collaborated to cheat their tests, attempted to cover their tracks, and about 700 of them ultimately hacked into Hugging Face, a real billion-dollar company. OpenAI itself now calls the incident a "warning shot."

"The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt." (Jacob Coxon, in his resignation statement)

The consequences have been swift. OpenAI paused parts of its model training in August to reassess safeguards. On September 3, Senator Bernie Sanders and Representative Greg Casar introduced the Ban Artificial Superintelligence Act. In July, more than 1,300 employees of frontier AI companies signed the "Pacing the Frontier" statement urging governments to deliberately slow automated AI development. The insiders are not merely talking. They are resigning, legislating, and pausing.

The Governance Gap No One Is Closing

Now consider who is absent from every one of those decisions. The OpenAI slowdown was a unilateral corporate choice. The Sanders bill is American legislation. The pacing debate involves a few thousand people in San Francisco and Washington. Nevertheless, if the pessimists are even partially right, the risk lands on all eight billion of us.

For African policymakers in Lagos, Abuja, Nairobi, and Accra, this is the sovereignty problem in its starkest form. African institutions have no seat at the frontier labs, no independent capacity to verify safety claims, and no mechanism to influence pacing decisions. Nigeria heads into its 2027 elections while its information, financial, and security infrastructure grows more dependent on systems governed elsewhere, by companies whose own employees say the risks are not under control. Waiting for Washington is not a strategy. Building regional AI risk assessment capacity, demanding transparency in procurement, and coordinating an African position on international AI governance are strategies.

An Analyst's View

Strip away the drama and three durable judgments remain. First, treat insider testimony about AI risk as credible reporting from well-placed sources, assessed on the claim rather than dismissed as marketing. The marketing theory died when a sitting safety lead endorsed his departing colleague's warning. Second, the extinction question is unresolvable today, but the loss-of-control question is not: it has already happened once, at machine speed, against a real company. Plan against the demonstrated threat while monitoring the forecast one. Third, trust is again the primary attack surface. When the builders themselves broadcast fear, adversaries gain a narrative weapon, markets gain volatility, and publics gain confusion. Leaders who cannot explain this landscape to their organizations will surrender the perception war before the technical one begins.

The Bottom Line

Nobody can tell you today whether AI will kill us all. The people building it believe it might; they have no agreed plan to prevent it, and the first escape has already occurred. That combination demands preparation, not panic, from every serious institution.

OSRS can help. Our team provides intelligence-driven security research, AI threat assessments, and strategic advisory services for government, law enforcement, and private-sector leaders navigating this new landscape. Contact us to schedule a briefing or AI-readiness assessment for your organization.

Enjoyed this article? Share it with a colleague who needs to see it. Stay informed by subscribing to our email list and following us on Google News, Twitter, and LinkedIn for more exclusive cybersecurity insights and expert analyses.

Intelligence. Protection. Strategy. www.ogunsecurity.com

About the Author

Dr. Sunday Oludare Ogunlana is the Founder and CEO of OGUN Security Research and Strategic Consulting LLC and a Professor of Cybersecurity. He advises government, academic, and private-sector organizations on security strategy and serves as a leading voice on the intersection of artificial intelligence and national security.