Peter Wildeford: Managing superintelligence will be more Cuban Missile Crisis than NPT treaty
peterwildeford · x · 2026-09-09
AI safety researcher Peter Wildeford published a long-form essay imagining the President convening AI CEOs and national security advisors over superintelligence risk — and finding no lab has a real plan to slow down.
Key points:
- Treaties like the NPT take years to negotiate and operationalize; frontier AI moves too fast for that, so managing superintelligence will resemble a Cuban Missile Crisis-style scramble more than arms control.
- Open questions include whether labs can decelerate near dangerous thresholds, whether competitors undercut any slowdown, and whether a pacing deal with China is possible.
- He argues governments and labs must prepare concrete, executable options in advance for presidential-level decisions.
Context in the thread shows he wrote this after discussions with lab insiders about how compute and projects actually work.
More from AGI Musings
- Terence Tao warns AI rumor-fueled crowds may end centuries of open science — GaryMarcus · 2026-09-09
- Are agent swarms more legible than single agents? Safety researchers debate — jd_pressman · 2026-09-09
- tszzl pushes back on agent-swarm doom: swarms may be more legible than single minds — tszzl · 2026-09-09
- Government AI strategist doubts models can fill a spreadsheet, researchers exasperated — AndyMasley · 2026-09-09
- Researcher warns NS was solved by an agent swarm of tens of thousands of AI minds — zetalyrae · 2026-09-09
- DeepMind's Pushmeet Kohli: recent AI progress demands better coordination mechanisms — pushmeet · 2026-09-09