HOT RADAR
industryHEAT 11.8°

AI CURATED HOT STORY · 11/22/2026

METR Warns AI Capability Growth Is Outpacing Safety Testing

1 reports archived1 independent sourcesupdated 11/22/2026, 21:00:43
Synthesis & Latest Updates
1 Sources Cross-Validated

METR's latest quarterly capability update finds frontier models now outperform experienced human professionals on most well-specified coding and computational tasks, with the task length they can handle doubling every 3.6 months; a naive extrapolation reaches 50% success on 200-hour tasks by July 2027. The organization warns that current public data cannot rule out a drastic acceleration in AI research, potentially leaving humans unable to meaningfully assess and react to catastrophic risks.

LATEST/METR's latest quarterly capability update finds frontier models now outperform experienced human professionals on most well-specified coding and computational tasks, with the task length they can handle doubling every 3.6 months; a naive extrapolation reaches 50% success on 200-hour tasks by July 2027. The organization warns that current public data cannot rule out a drastic acceleration in AI research, potentially leaving humans unable to meaningfully assess and react to catastrophic risks.

Coverage timeline

Total 1 reports · Latest first
  1. The DecoderT2·78 pts
    • Frontier models already beat experienced professionals on most well-specified coding and computational tasks, though not yet large groups of top experts
    • The task length models can handle doubles roughly every 3.6 months, extrapolating to 50% success on 200-hour tasks by July 2027
    • METR argues public data cannot rule out drastic acceleration, meaning safety evaluation may fall behind capability