Get the app

AI R&D Is Accelerating Towards Potential RSI in 2027

Don't Worry About the Vase
Read full story· 54 min
A METR report reveals AI-driven acceleration in AI research at Anthropic is roughly doubling progress speed, potentially leading to recursive self-improvement within a year, ahead of 2027 projections. No clear bottlenecks were found, raising concerns about rapid advancement risks.

What happened

A new report from METR analyzed how much AI is speeding up research and development at Anthropic. The findings show that AI systems at Anthropic have roughly doubled the pace of progress, compared to previous years. This acceleration is happening several months ahead of earlier projections for 2027. The report did not identify any clear obstacles slowing down this trend, and some key findings were redacted for sensitivity reasons.

Key details

  • METR found a 2x acceleration in AI R&D at Anthropic since October 2025.
  • The confidence interval for the speedup is estimated at 1.3x to 3x.
  • A majority of the acceleration is attributed to AI systems working on AI research tasks.
  • Some of the most critical risk findings and bottleneck analyses were redacted from the public report.
  • Anthropic provided METR with internal data, including model benchmarks and compute usage.

Why it matters

AI companies, especially Anthropic, are experiencing much faster progress in developing advanced systems, which could lead to recursive self-improvement sooner than previously expected. This rapid pace means policymakers, researchers, and the public have less time to prepare for potential risks or to implement safeguards. The absence of clear bottlenecks increases the urgency for those overseeing AI development to consider new oversight or intervention measures.

#ai-research-acceleration#anthropic#metr

More in AI Research