Extrapolating Anthropic's Risk Report: RSI a Lower Bound Before Late 2026
peterwildeford · x · 2026-08-25
Saif Khan extrapolates from Anthropic's August 2026 risk report that fully automated AI R&D (RSI) may arrive between Dec 2026 and Feb 2027 (or 2028 under pessimistic assumptions). He regresses CoBench (Anthropic's internal automated AI R&D benchmark) against AECI scores across recent Claude models, using the report's claim that a model fully substituting for Anthropic research staff should score at least 85% on CoBench.
Peter Wildeford argues (and Khan has conceded) the correct reading is that RSI is highly unlikely before Sep 2026 — the benchmark is a lower bound on RSI, not evidence it lands around Dec 2026. After Dec 2026 it becomes harder to rule out RSI, though he still considers it unlikely.
Related event: Analysis Suggests Fully Automated AI R&D May Arrive by Late 2026(2 posts)→
More from AGI Musings
- AI-driven job losses could turn $10T consumer debt into a $1T banking hole — churchkey · 2026-10-16
- Into the Unknown: an AI agent benchmark of 15 open research problems from US national lab scientists — Jeande_d · 2026-10-16
- Soon your AI will meet their AI — and humans without agents can't compete — danfaggella · 2026-10-16
- Analyst: in a year, models 2-3x better across the board at 10x lower cost, token use up 10-20x — DeryaTR_ · 2026-10-16
- AI will make many discoveries for us, but adversarial domains like soccer will thrive — natanielruizg · 2026-10-16
- MIT's Buehler on AI as a new language of science: superintelligent partners and executable papers — CatAstro_Piyush · 2026-10-16