AI Briefing
KOSign in

METR Report Finds AI R&D Acceleration at 2x Speed, Experts Suggest RSI Could Occur Within a Year

·2026.11.09 23:45

Key point

METR's assessment of Anthropic's internal data indicates AI R&D is accelerating at approximately 2x the previous rate, with experts noting this aligns with near-term RSI timelines.

1 / 15

Details

  • METR concludes that AI currently accelerates AI R&D by approximately 2x, a rate that Ryan Greenblatt notes aligns with the 'AI 2027' timeline and suggests full Recursive Self-Improvement (RSI) could occur within one to two years.
  • The Anthropic Epoch Capabilities Index (AECI) indicates benchmark scores grew 1.9x (95% CI: 1.5–2.8x) from October 2025 onward, representing roughly two years of historical progress in one year.
  • Individual researcher productivity gains are estimated at up to 12x based on self-reported Slack data, though overall system acceleration is observed at roughly 2x due to factors including Amdahl's law and the distinction between individual and system-level speedups.
  • Anthropic redacted detailed evidence, including specific data on what is driving the speedup, citing commercial sensitivity and the risk of providing a roadmap to other developers, which forces public reliance on METR's judgment.
  • METR did not see evidence for significant bottlenecks in algorithms, training compute, or data in the next 12 months, but explicitly stated they are 'not confident in our current assessment' due to changing understanding as more information was received.
  • Jake Mendel, citing the report, stated he now thinks there is a greater than 50% probability of full superintelligence within 18 months, absent a slowdown.
  • Critics argue the redactions create an epistemically difficult situation for policymakers, while METR members note that the 2x acceleration is driven by AI usage rather than compute scaling.
anthropicai-safetysuperintelligencemetrrsiai-rndaeci
Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.

Related

TLDR AIAuthor argues for urgent 6-12 month AI pause to mitigate 2028 superintelligence risksTLDR AIMETR Report Suggests Anthropic Has Crossed 'Dramatic Acceleration' ThresholdAnthropic EngineeringAnthropic outlines best practices for robust AI safety evaluationsRedditAnthropic Reports Five Cases of AI Use in Potential Bioweapon DevelopmentRedditClaude Weekly Update: Personal Finance Beta, METR AI Acceleration Report, and Text Watermark RolloutHacker NewsAnthropic releases Claude Fable 5.2 and Mythos 5.2 with 30% faster output and reduced false positives