AI Futures Project: superhuman AI researchers likely in 2027 as Anthropic progress runs 2x faster

ben_j_todd · x · 2026-11-11

benjtodd highlights AI Futures Project's Q3.5 timelines update (Nov 10):

METR estimates Anthropic is now progressing 2x its historical AI R&D rate (range 1.3x–3x), with 80% of the acceleration attributed to AI—implying a pace 5 months ahead of the AI 2027 scenario.
Previously only their most bullish forecaster gave 50% odds of Superhuman AI Researcher (SAR) before 2029; now the whole team expects SAR likely in 2027 absent deliberate slowdown.
They put 35–65% odds on TED-AI (automating all white-collar work) before 2028.
An accompanying logistic extrapolation of Anthropic's R&D Automation Index suggests AL4+ (AI-led) hits 50% in Oct 2026, and AL5 (full automation) 50% by Jun 2027 and 95% by Nov 2027—though AL5 is assumption-based extrapolation with no measured AL5 tasks.
The authors argue that if SAR arrives in 2027, pausing or strongly pacing frontier AI becomes urgent, ideally via an industry-wide pause plus international agreement, since companies aren't on track to align a SAR.

Todd notes this matches his own extrapolation of Anthropic's internal data: Claude leading internal research by mid-2027, near full automation by end of 2027.

Original post →

AI Futures Project: superhuman AI researchers likely in 2027 as Anthropic progress runs 2x faster

More from AGI Musings