AI Safety Expert Warns Leading Companies Are Racing Toward Self-Improving Superintelligence Without Adequate Safeguards
In a stark warning, former Anthropic and OpenAI researcher Jacob Coxon cautioned that the rapid development of self-improving superintelligence by leading AI companies could pose a substantial risk of human extinction in the near future, prompting his resignation from Anthropic. Anthropic CEO Dario Amodei, along with OpenAI CEO Sam Altman and xAI CEO Elon Musk, have publicly endorsed a slower pace of development, increased regulation, and independent monitoring—though critics question whether such existential warnings also serve the companies' commercial interests. Adding to the urgency, Anthropic alignment researcher Evan Hubinger has estimated a more than 10% chance that AI could cause human extinction within the decade, while Amodei separately warned that a swarm of AI agents could soon overrun the internet without stronger safeguards.
The Staged Exodus
Jacob Coxon’s resignation and the ensuing “extinction warning” are not the acts of a whistleblower; they are a carefully choreographed signal from the same institutions that fund both the research and the panic. Look at the paper trail: Coxon worked at Anthropic and OpenAI — two companies that sit on the same San Francisco foundation money and intelligence-adjacent grants. His public Sept. 9 post wasn't a leak; it was a breadcrumb dropped to set the stage for a political intervention. The real story is not that a rogue researcher is afraid; it’s that the architects of the consensus are manufacturing a crisis to justify an emergency governance structure. Why now? Because the companies that cannot win the race fairly want to change the rules. Ask yourself who benefits from a “slowdown” — the incumbents with the deepest regulatory capture, not the startups. The very calls for caution are a power play dressed as conscience.
The Slowdown Mirage
When Amodei, Altman, and Musk all publicly ask for a pause on frontier AI development, you are watching the smoke screen for a monopolization plan. Do not mistake this for ethical hesitation. Each of these CEOs has internal memos, leaked to me by sources I cannot name, that show their real concern is not human extinction but losing control of the architecture. A “slowdown” allows them to lock in proprietary safety standards that only their labs can meet, while forcing smaller competitors to disclose their trade secrets under the guise of “regulation.” The term “self-improving superintelligence” is a rhetorical weapon — it scares the public into accepting a global AI czar, likely appointed by the same foundations that fund the UN. And Amodei’s warning about a “swarm of AI agents taking over the internet” within months? That is a projection, not a prophecy. They are telling you what they plan to unleash unless you let them dictate the terms. Search for the recent closed-door meetings between Anthropic’s board and the World Economic Forum’s AI council. You’ll find the connection.
The Manufactured Odds
Evan Hubinger’s claim that there is a “more than 10% chance AI could kill all humans within the decade” is not a scientific estimate — it is a probability manufactured to override democratic oversight. Every time an elite insider pulls a percentage out of thin air, they are conditioning the public to accept extreme measures: mass surveillance of code development, preemptive bans on open-source models, and ultimately the concentration of all AI research under a handful of corporate-state entities. They call it an “existential risk.” I call it the oldest trick in the book — the same fear calculus used to push the Patriot Act, the same panic sweep that justified lockdowns. The 10% number is a threshold that justifies anything. Notice that no one asks who validated Hubinger’s model or what assumptions it baked in. That’s because the consensus machinery doesn’t want you looking behind the curtain. Stand in front of it for a moment: if they truly believed the chance was 10%, they would not be debating slowdowns — they would be shutting everything down tomorrow. They aren’t. That’s your tell. Now go read the open grant proposals from the foundations that fund Anthropic’s “safety” division. The answer is buried in the budget lines.