RESEARCH·Nov 9, 2026, 4:26 PM ET

Ajeya Cotra cites METR assessment showing Anthropic's capability gains sped up

TL;DR:

  • Ajeya Cotra's account of a METR assessment says Anthropic's measured capability progress ran roughly twice as fast over the latest year.
  • METR investigated how Anthropic's own AI systems are accelerating Anthropic's research, using internal evidence.
  • Ajeya Cotra argues that the measured acceleration meets a threshold in Anthropic's safety policy.
  • Substantial redactions limit public scrutiny of the supporting evidence.

Summary

METR investigated how Anthropic's own AI systems are accelerating its research, with access to internal evidence. Cotra's reading of the assessment identifies roughly a twofold increase in the pace of measured capability gains. She argues that this meets an acceleration threshold in Anthropic's safety policy, but substantial redactions limit public scrutiny. Her conditional extrapolation to superintelligence within a year isn't established by the reported measurement.

Open planned-obsolescence.org in a new tab ↗
Powered by Moatly Lens

The Bigger Picture

The headline tells you what happened. Discover the facts behind it, the forces at work, the connections that matter, and what comes next.

Why This Matters

The speed of AI development has hit a massive gear shift. Recent evaluations show progress on complex tasks is moving twice as fast as last year, with internal tools now handling high-level physics and math research. This acceleration matters because it forces a collision between rapid technical gains and the safety policies designed to slow them down.

What We Actually Know

  • Anthropic launched its Mythos and Fable models following an external assessment of its internal research speed.
  • Evaluators found that the company's progress on well-defined tasks is now running at roughly twice the rate seen a year ago.
  • Internal AI systems like Fable 5.2 are already contributing directly to complex research in physics and mathematics.

How the Pieces Connect

  • The nonprofit METR has a history of evaluating model capabilities for major labs including OpenAI and Anthropic.
  • This acceleration occurred while the company's CEO publicly called for the industry to slow down for safety reasons.

What This Could Change

  • InferenceThe current pace suggests the company has completed two years' worth of historical progress in just the last twelve months.
  • ScenarioIf these gains meet safety policy thresholds, the company may be required to implement stricter oversight or pause certain developments.

Where the Story Goes Next

  • How much of the reported productivity gain is due to AI versus improved human workflows or better management?
  • Will the redacted portions of the METR report be made public to allow for broader scientific scrutiny?