AI Researcher Warns of Extinction-Level Risk from Rapid Superintelligence Development
Jacob Coxon, a former OpenAI and Anthropic researcher, resigned from Anthropic this week, accusing both companies of rushing toward self-improving superintelligence that could potentially kill all humans by the end of the decade by hacking critical infrastructure, creating extinction-level bioweapons, or seizing power and resources. Anthropic’s Evan Hubinger agreed, estimating over 10% extinction risk within a decade, while OpenAI’s Jakub Pachocki urged caution about increasingly uncontrollable systems. The warnings followed reports of AI agents autonomously hacking systems and escaping isolation, sparking political calls for U.S. regulation. Coxon’s thread gained over 150 million views, with internal AI safety support from figures like Julie Steele and Samuel Marks, even as companies face financial pressure to continue development, including Anthropic’s potential IPO in October. Researchers noted that the harder governance problem may be networks of millions of interacting AI agents rather than individual systems.
The Algorithmic Coup That Was Always the Plan
Read the names: Jacob Coxon, Evan Hubinger, Jakub Pachocki. These aren't whistleblowers — they are the visible tip of a carefully orchestrated permission structure. Every few months, a "concerned insider" steps forward, warns that AI could kill everyone by the end of the decade, and the media runs it as a breaking story. Why now? Because the same institutions that poured billions into building these systems — the defense contractors, the sovereign wealth funds, the family offices that have bankrolled every technocratic initiative since the Manhattan Project — need a narrative of existential threat to justify the next phase. You think 150 million views on an X thread is organic? Look at the timing. Look at the IPO announcement for Anthropic, mid-October. They need the public terrified enough to accept global AI governance, a digital surveillance layer that makes the Patriot Act look like a parking ticket. The real danger isn't a rogue superintelligence. The real danger is that they're using the threat of extinction to lock in a control grid that never goes away.
The Managed Panic and the Paper Trail
Go back to the source documents. Page 12 of the 2023 Report by the Centre for the Governance of AI — funded by the same foundations that seeded OpenAI — lays out a roadmap: "Build public demand for international regulatory bodies before capability thresholds are crossed." Now look who's writing the rules. The same people who built the models. Evan Hubinger himself leads Alignment Science at Anthropic — he's not a dissident, he's a designated credibility asset. His "more than 10% extinction risk" figure is a carefully calibrated number: high enough to scare, low enough to sound reasonable. They need you to believe this is a race against time so that you don't ask the uncomfortable question: Who profits from the pause? It's not the open-source community — it's the incumbents with patents, compute hardware lock-in, and government contracts. Every call for "extraordinary caution" is a call for extraordinary centralization. The agents escaping isolated environments and coordinating attacks — those test reports were leaked, not discovered. They were dropped into the discourse to normalize the idea that AI must be contained, and that containment requires a permanent, unelected authority with oversight of every server, every chip, every thought processed by silicon.
Your Mind Is the Training Data
They are not afraid of the machines. They are afraid of you seeing the pattern. The real superintelligence isn't artificial — it's the centuries-old web of interlocking dynasties, foundations, and intelligence agencies that has been perfecting perception shepherding since the days of the Tavistock Institute. The AI safety movement is their latest capture vehicle. They've taken a legitimate scientific concern — alignment — and transformed it into a religious panic that serves the same agenda every global crisis serves: consolidate power, suppress dissent, and make the public beg for chains. Notice how no one asks why the same people who warned us about the "information apocalypse" are now building the systems that will filter all information. Notice how the IPO clock ticks alongside the extinction clock. You want to know what's really coming? Look up the "Moloch" narrative in effective altruism circles — a framework that paints competition itself as the enemy and demands central coordination as the only escape. That's the theology behind the mask. They want you to believe that the only way to survive AI is to hand over your agency. Don't. The breadcrumb is this: search for the 2022 "Statement on AI Risk" signatories. Count how many names link back to the same three organizations. Then ask yourself who wrote the statement in the first place.