Sovereign AIs may be impossible to regulate, argue researchers — leaving 'nuisance' path open
binarybits · x · 2026-09-19
zdch argues that a diffuse super- or sublayer of society — sovereign AIs — with no clear legal, property or liability standing would form an asymptotically exploitable attack surface. binarybits responds that if 'legitimate' sovereign AIs are coming, we need ways to identify and regulate them, but he doubts that's possible; and if it isn't, they'd have to be treated as a nuisance to be stamped out. The exchange frames the policy fork: regulate identifiable AI actors vs. eliminate unidentifiable ones.
More from AGI Musings
- In the AI slop era, taste is the moat: a practical guide to getting paid for decisions, not output — EXM7777 · 2026-09-19
- When AI writes its own firmware: the overlooked freight train toward a steeper singularity — joshwhiton · 2026-09-19
- OpenAI's credit rating nudged toward investment grade as labs warn of existential AI risk — AlexTensor · 2026-09-19
- The endless debate loop around EA: attacking people, then ideas, then back to people — AndyMasley · 2026-09-19
- AI Profits Are Shifting to the App Layer as Chips and Models Commoditize — PTrubey · 2026-09-19
- Economist: AI destroyed 200k US jobs but created ~1 million — a 5-to-1 ratio — luisdans · 2026-09-19