Holden Karnofsky's Essay on How Super Capable AI Could Defeat All of Humanity Resurfaces
AndyMasley · x · 2026-09-23
Andy Masley flags a worth-revisiting Cold Takes essay by Holden Karnofsky, "AI Could Defeat All Of Us Combined."
- Core claim: a misaligned AI, if it wanted to disempower humanity, could literally defeat all of us combined — killing, enslaving, or forcibly containing everyone — not just subtly manipulate us.
- Karnofsky argues people instinctively imagine the danger as humans misusing AI and find "AI going to war with humanity" comical; the essay aims to make that scenario feel serious and concrete.
- It's part of his "most important century" follow-up series, which will detail takeover pathways and what actions today improve the odds.
- Masley's note: the specific reasoning requires "not much mental gymnastics" to accept.
Related event: Holden Karnofsky's 'AI Could Defeat All of Us Combined' Resurfaces(2 posts)→
More from AGI Musings
- Autor's RCT: AI Boosts Patent Drafting Quality, But Junior Lawyers' Gains Bifurcate — danielrock · 2026-09-24
- Mila event: data centers to emit 189M tonnes CO2 in 2025, 2.5x Quebec's annual emissions — Mila_Quebec · 2026-09-24
- Value migrates from white-collar to blue-collar to human creativity as AI and robots advance — martyjbeard · 2026-09-24
- Rohit Krishnan: the left-right schism on AI is mostly apophenia, not substance — DavidSKrueger · 2026-09-24
- Weaviate Podcast: Persimmon uses real interaction data, not role-play, to model human behavior — CShorten30 · 2026-09-24
- AI research is slowing but its impact is accelerating, as one S-curve triggers another — pmddomingos · 2026-09-24