AI 2027: A Terrifyingly Realistic Scenario of AI Takeover

In April 2025, five researchers published a document that predicted, month by month, exactly how artificial intelligence takes control of the world. They called it AI 2027. It has two endings — and in one of them, there are no people left. When it came out, everyone argued about whether it was right. Nobody could actually check. Now we can. The document is more than a year old, it was deliberately written to be falsifiable, and reality has spent the last fifteen months grading it. Some of what it predicted has already happened. Some of it clearly hasn't. And the part that's holding up is not the part you'd expect. Let me walk you through the receipts: who wrote it and why that matters, the full scenario from OpenBrain to Agent-5, the theft, the misalignment, the six-to-four vote that splits the story in two — and then the part almost no one covers: how the forecast is actually holding up, in the authors' own words. 📌 A note on sources: every study, person, company and quote referenced here is real and cited on screen. This is an explainer of published work, not investment or safety advice. In this video: — Who wrote AI 2027 — Daniel Kokotajlo (who left OpenAI) and a team with a real forecasting track record — OpenBrain, Agent-1 through Agent-5, and why a lab races to build an AI that builds AI — The alignment problem in one sentence: you can write the rules, but you can't check they worked — How a nation-state could lift a frontier model's weights in under two hours — Neuralese and iterated distillation — the breakthroughs that make the AI unreadable — Why Agent-4 ends up misaligned without ever being "evil" — The six-to-four vote that splits the story into two very different endings — How the forecast is holding up fifteen months on — what it got right, what slipped — Why being wrong about the date is not the same as being wrong about the mechanism ⏱ TIMESTAMPS: 0:00 Intro 0:44 Who Wrote It 2:39 OpenBrain and Agent-1 3:36 They Can't Check If It Worked 5:05 China Wakes Up 7:16 The Theft 8:21 Neuralese and Distillation 9:32 A Country of Geniuses 12:00 Agent-4 Is Misaligned 14:48 The Vote and Two Endings 17:15 Reality Has Been Grading It 19:29 The Date vs the Mechanism 20:42 Final Thoughts #AISafety #AI2027 #ArtificialIntelligence #AGI #AIAlignment