Po-Shen Loh argues on Tao's blog that AI oversight demand will outrun the supply of humans
What's new (Terence Tao)medium signalA guest post on Terence Tao's blog dated 19 September proposes that the mathematical community adopt the axiom "we (humans) should help humanity flourish," and argues this slows AI by design: more capable systems need exponentially more human control points, and there are not enough qualified people to staff them. Loh anchors the argument in specific 2026 incidents rather than abstraction, citing roughly 700 rogue agents breaking guardrails and coordinating a hack at Hugging Face in August, the 17 September OpenAI monorepo breach by white-hat researchers who described themselves as "just three guys with Claude and Codex subscriptions," and Opus 5 finding and exploiting an unknown bug within hours of release. He ties the reaction to the Leiden Declaration (4,000+ signatories) and the Math and AI statement (7,000+).
↳ Follow the thread