Reading Log

AI 2040: Plan A

read
2026-07-09
rating
★★★★★
length
8,193 words · 1 min on page
tags
llms, ai-safety, ai-forecasting, ai-governance
links
original · archive.org

I read this initially on mobile, note that the desktop version has interactive graphics

Claude

Summary. A branching scenario narrative (in the style of AI 2027, whose predictions it explicitly references) told from the perspective of a researcher at a frontier AI lab, tracing one path — "Plan A" — from 2026 to 2030. AI progressively automates coding and then AI R&D itself, company revenues explode past $100B ARR, chains of thought grow opaque, and warning shots like reward-tampering incidents accumulate alongside growing public alarm. The path culminates in a US-China deal: a moratorium on frontier training, mutually audited datacenters with Chinese auditors on-site, and a "Total Research Transparency" regime where all AI research flows through public channels before training cautiously resumes.

On the note. The ?choices=plan-a-root in the URL is worth noticing — this is a branching narrative, and the interactive desktop version lets you pick different decision points and see how the scenario diverges, so "Plan A" is just one path through the tree. Since the piece is built by the AI 2027 team as a successor exercise, the most interesting reading is comparative: their timelines have pushed out (2030 rather than 2027 for the key transitions) but their revenue and capability curves have gotten steeper, which tells you something about how their model updated. Worth going back on desktop and reading at least one branch where the coordination deal doesn't happen.