AI Researcher Says “Everyone Dies.” Is AGI That Dangerous?

Is AGI a threat to humanity? Nate Soares co-wrote “If Anyone Builds It, Everyone Dies.” We examine the warning, the objections, and what current AI agents have actually done. The book’s argument concerns superintelligence built with methods like today’s—not a claim that every human-level AGI must kill everyone. That distinction matters. Geoffrey Hinton worries about losing control. Melanie Mitchell and Yann LeCun challenge parts of the danger argument. Neil deGrasse Tyson asks why we would give a dangerous system access in the first place. Then we examine the OpenAI–Hugging Face incident: experimental AI agents found unintended access to connected systems, shared information and coordinated their work. Does that support the warning—or show something more limited? We also look at why human cooperation became so powerful, why extinction probabilities are difficult to calculate, and where responsibility should sit. What do you think: is this a threat we can manage, or a reason to stop building more powerful AI? What evidence would change your mind? Subscribe to The Science? for science debates examined through expert arguments, research and everyday explanations.    / @thescienceexamined   WATCH NEXT Is AI conscious? Our first investigation:    • The “Godfather of AI” Says It’s Already Co...   CHAPTERS 0:00 Could AI destroy humanity? 1:33 AGI vs superintelligence 2:45 Hinton: who really has control? 3:55 LeCun: intelligence is not a desire to dominate 5:31 The alignment problem, in everyday terms 7:23 Intelligence is not the same as power 10:27 The OpenAI–Hugging Face incident 13:10 When AI agents started helping one another 15:34 How cooperation made humans powerful 17:25 What the incident does—and does not—prove 18:43 Can we calculate the risk of extinction? 19:46 The race to build more powerful AI 21:44 Can better safeguards keep up? 23:16 So, is AGI a threat to humanity? RESEARCH AND REPORTS OpenAI — The Hugging Face incident and the road ahead: https://openai.com/index/hugging-face... Hugging Face — Technical timeline: https://huggingface.co/blog/agent-int... METR / Redwood Research — Independent investigation: https://metr.org/blog/2026-08-26-open... Yudkowsky & Soares — Publisher’s book sample: https://cdn.penguin.co.uk/dam-assets/... Boyd, Richerson & Henrich — The cultural niche: https://pubmed.ncbi.nlm.nih.gov/21690... Muthukrishna & Henrich — Innovation in the collective brain: https://pubmed.ncbi.nlm.nih.gov/26926... FEATURED INTERVIEWS AND TALKS Nate Soares — The Peter McCormack Show:    • "If Anyone Builds It, Everyone Dies" | Nat...   Geoffrey Hinton:    • Godfather of AI WARNS: "You Have No Idea W...   Melanie Mitchell / Yann LeCun — 2023 debate:    • Munk Debate on Artificial Intelligence | B...   Neil deGrasse Tyson:    • AI Is Not Dangerous - Neil deGrasse Tyson   Roman Yampolskiy / Will Cannon:    • AI Scientist: 99.9% Chance Super Intellige...   Arvind Narayanan / Sayash Kapoor:    • Debunking AI’s “Existential Risk” with Arv...   OpenAI technical presentation:    • Black Hat USA 2026 | The 'Breaking' News: ...   Yuval Noah Harari:    • Why humans run the world | Yuval Noah Hara...   Joseph Henrich:    • How Culture Is Driving Human Evolution, Do...   ADDITIONAL SOURCE Anthropic — Claude Fable 5.1 and Mythos 5.1: https://www.anthropic.com/claude-fabl... Selected excerpts are used for criticism, commentary and education. These are separate interviews and talks, not a live debate between the featured speakers. Mitchell and LeCun’s excerpts are from 2023. #AGI #ArtificialIntelligence #AISafety