← AI news, last 7 days

METR report on Anthropic AI R&D acceleration

Last scan Sat Nov 21, 06:54 UTC
6 sources · tracked since Nov 6 · 7 new articles this week · 2 independent origins: METR, MIT Technology Review

METR, an AI auditor, produced a report with heavy redactions suggesting that AI could be making Anthropic's AI progress go twice as fast, and Platformer characterizes its conclusion as being that the company's speed of AI research and development had approximately doubled.12 Eli Lifland sums up METR's figure as close to 2x. He says METR wouldn't be shocked by a value anywhere between 1.3x and 3x, and that it speculated roughly 80% of today's speedup stems from AI.3 Lifland says his median for a Superhuman AI Researcher moved from about 2030 to 2027, and Daniel Kokotajlo says that, if METR is right, the result implies an AI-driven acceleration 5 months ahead of the AI 2027 scenario.45

Paul Christiano says the data are consistent with AI progress having roughly doubled over the last 6-12 months, and he sees a reasonable chance of very rapid recursive-self-improvement-driven progress in that window.67 Evidence quality is a point of complaint, though: Christiano says most of what the public has is METR and Anthropic asserting their conclusions, since the full report was redacted.89

Dwarkesh Patel notes the measured doubling concerns verifiable tasks and asks whether progress on non-verifiable tasks has also doubled.1011 Zvi Mowshowitz calls it the "null METR report" and puts the probability of full AI R&D automation in 2027 under 50%.1213

In a separate METR update, models fell short on judgment-heavy tasks: none of the AI-designed evaluations was judged useful by METR employees, though newer models like Opus 5.5 reached "borderline above the bar".1415

The report landed as Anthropic went public, and critics seized on it: Platformer says the doubling would indicate Anthropic passed its own "Automated R&D" threshold, and the Future of Life Institute cited it in a statement condemning Anthropic's pursuit of recursive self-improvement. inference1617

Every sentence was checked word for word against quotes from the articles it cites. Each footnote shows its quote, unless the page already quotes as much of that article as we allow; then it links the article instead. A model wrote and arranged the sentences, and a second read checked each against its quotes (written 2026-11-20).

Sources quoted

  1. A recent report from AI auditor METR concluded that Anthropic’s rate of AI R&D progress had roughly doubled due to AI’s research ability. — Platformer, Anthropic goes public
  2. a different, heavily redacted METR report from earlier this month, which claimed that AI might be speeding up the pace of AI progress at Anthropic by a factor of two — MIT Technology Review, The most misunderstood graph in AI just got a major update
  3. METR recently estimated that Anthropic is now making AI progress about twice as fast as its historical rate (they would not be surprised by anything from 1.3x to 3x), and guessed that “perhaps ≈80%” of the current acceleration is due to AI. — Zvi Mowshowitz, Quickly, There’s No Time
  4. My previous median for Superhuman AI Researcher was ~2030, now it’s 2027. — Zvi Mowshowitz, Quickly, There’s No Time
  5. that implies an AI-driven acceleration that is 5 months ahead of the AI 2027 scenario — Zvi Mowshowitz, Quickly, There’s No Time
  6. we are seeing data consistent with the pace of AI progress having roughly doubled over the last 6-12 months — Dwarkesh Patel, Paul Christiano – We could be 12 months away from AIs that can overthrow the government
  7. I do think the probability of seeing that kind of extremely rapid progress driven by recursive self-improvement over the next 6-12 months is reasonably high. — Dwarkesh Patel, Paul Christiano – We could be 12 months away from AIs that can overthrow the government
  8. METR saying, “We had some conclusions. We promise we have a full report, but the entire report has been redacted.” — Dwarkesh Patel, Paul Christiano – We could be 12 months away from AIs that can overthrow the government
  9. Mostly we just have them saying that, which is not a very good situation. — Dwarkesh Patel, Paul Christiano – We could be 12 months away from AIs that can overthrow the government
  10. Progress on verifiable tasks has doubled. — Dwarkesh Patel, Paul Christiano – We could be 12 months away from AIs that can overthrow the government
  11. So you can ask the question, has progress on non-verifiable tasks also doubled over the last year? — Dwarkesh Patel, Paul Christiano – We could be 12 months away from AIs that can overthrow the government
  12. It was the null METR report. — Zvi Mowshowitz, On Dwarkesh Patel’s November 2026 Podcast With Paul Christiano
  13. This is why I continue to believe that the probability of full automation of AI R&D in 2027 is under 50%. — Zvi Mowshowitz, Quickly, There’s No Time
  14. Not one of the AI-designed evaluations was good enough for METR employees to consider it useful for their research. — MIT Technology Review, The most misunderstood graph in AI just got a major update
  15. Some of the newest models, such as Opus 5.5, tend to achieve what METR calls “borderline above the bar” performance on the evaluation design tasks — MIT Technology Review, The most misunderstood graph in AI just got a major update
  16. doubling the rate of research progress would indicate that the company had surpassed its own risk threshold for “Automated R&D” in its “Responsible Scaling Policy.” — Platformer, Anthropic goes public
  17. An independent report from METR provided evidence that Anthropic’s pursuit of RSI has dramatically sped up AI development – by a factor of nearly 2x. — Future of Life Institute, Leading figures condemn Anthropic for reckless race to recursive self-improvement

Who reported it

Facts and takes

No checked facts or takes on this story yet.

Timeline

Built 2026-11-21 06:54 UTC. Facts and quotes say who states them, not that they are true.