AI Risk Discourse Has Moved From Heuristic Arguments to Specific Takeover Threat Models
herbiebradley · x · 2026-09-24
Herbie Bradley revisits Holden Karnofsky's essay spelling out how extremely capable AI could take over, calling it emblematic of the 2022-era AI risk discourse he found unconvincing—built on high-level heuristic arguments.
By contrast, he argues today's much better understanding of reward-seeking behavior and cyber-offensive AI capability allows spelling out concrete takeover threat models rather than abstract speculation.
Related event: Holden Karnofsky's 'AI Could Defeat All of Us Combined' Resurfaces(2 posts)→
More from AGI Musings
- Philosopher Peter Godfrey-Smith on non-human AI consciousness and existential risk — AnnaCiaunica · 2026-09-24
- Mat Dryhurst calls out viral reposted quotes as AI slop with bot applause — matdryhurst · 2026-09-24
- Aesona's founding thesis: your 'second self' AI model must be owned by you — aaron_lou · 2026-09-24
- Is the market pricing that ~100% of enterprise data will flow through LLMs in 3 years? — gabriel1 · 2026-09-24
- Gallup: positive feelings toward AI outweigh negative in 34 of 37 countries, yet 57% have never used it — rohanpaul_ai · 2026-09-24
- New paper uses multiscale NeuroAI model to explain the zolpidem consciousness paradox — introspection · 2026-09-24