If Firms Can't Stop China Stealing Weights, They Shouldn't Build Risky AI

trevposts · x · 2026-11-18

Quoting a post on pacing AI (capping the share of compute used for AI R&D; not advancing capabilities until you can prove China couldn't steal your weights), the author argues: since companies are far from China-proof on security, they must not develop AI systems that, if stolen by China, would pose a major risk to US national security — including by making it impossible to coordinate with China to ban full recursive self-improvement (RSI).

Original post →

More from AGI Musings

AGI Musings channel →