METR finds AI has roughly doubled Anthropic's AI R&D pace, key evidence redacted
METR released an independent evaluation of Anthropic: since August, the org has used Anthropic's internal information to study "whether AI is accelerating AI R&D itself," concluding that Anthropic's benchmark progress has sped up roughly 1.9–2x, and that a large share of the gains may be AI-driven. Many commentators see this as potentially one of the biggest turning points in the history of technology, with implications far beyond chatbots. The report has also drawn transparency concerns due to heavy redactions.
Confirmed
- METR officially published the report, estimating about a 1.9x acceleration in Anthropic's benchmark progress (@thione), with the top-line conclusion framed as up to 2x (@austinc3301, @iruletheworldmo).
- Part of the acceleration comes from AI participating in AI research work, i.e., models accelerating the iteration of model capabilities.
- Chart info: frontier model evaluations improve 15.8 points per year across labs, with the latest Claude Fable 5.2 jumping +4.9 points in a single step (@iruletheworldmo).
- The evaluation targets Anthropic's progress on recursive self-improvement (RSI); Anthropic is known to expect RSI to arrive soon (@trevposts).
- Large portions of the report were redacted at Anthropic's request on "commercially sensitive" grounds, removing evidence METR considered most informative (@thione, @trevposts, @austinc3301).
Unconfirmed
- The original report and full reasoning cannot be publicly scrutinized; the derivation details behind numbers like 2x/1.9x are unverifiable due to the redactions.
- The post relayed by @m2 gives no specific data from the evaluation, and its "turning point" framing is commentary rather than report content.
Why it matters
- @trevposts notes that both Ajeya Cotra and Ryan Greenblatt treat this report as a major update toward shorter AI timelines, sparking heated debate in timeline-forecasting circles.
- @trevposts has called on the US Congress to subpoena the full report, highlighting the tension between safety evaluation and commercial secrecy.
- If AI is indeed already accelerating AI R&D, capability gains may be entering a self-reinforcing loop, making the pace of AI progress itself a variable that demands close monitoring.
2026-11-08 ~ 2026-11-10 · 9 related posts
- Episode 1: Anthropic Says It May Imminently Hit RSP Threshold for AI-Accelerated R&D(2026-11-03, 7 posts)
- Episode 2: AI 2027 Authors Warn Anthropic's 2x Threshold Could Accelerate Superintelligence Timeline(2026-11-05, 2 posts)
- Episode 3: METR Report: Anthropic AI R&D Speeds Up 1.9x, RSI Expectations Shift Sharply(2026-11-06, 145 posts)
- Episode 4: Report: Anthropic's Next Model Finished Training, Tops All METR Benchmarks(2026-11-06, 2 posts)
- Episode 5: METR Says Anthropic Canceled Compute Allocation Assessment Pilot, Sparking Criticism(2026-11-06, 4 posts)
- Episode 6: AI Safety Community Criticizes Anthropic's Acceleration Strategy(2026-11-06, 2 posts)
- Episode 7: Oxford philosopher explains why AI benchmark error bars have widened(2026-11-06, 2 posts)
- Episode 8: Anthropic's Internal Models Are Ahead of the AI 2027 Forecast Curve(2026-11-07, 3 posts)
- Episode 9: Anthropic's 10x AI Research Speedup Debated(2026-11-08, 2 posts)
- Episode 10: METR Data Shows AI Progress Accelerating While Returns per Billion Dollars Shrink(2026-11-08, 9 posts)
- Episode 11: METR finds AI has roughly doubled Anthropic's AI R&D pace, key evidence redacted(2026-11-08, 9 posts)
- Episode 12: Debate: Is Accelerating Individual Researchers a Useful Metric?(2026-11-09, 2 posts)
- Episode 13: AI Acceleration Challenges RL Catch-Up Hypothesis(2026-11-09, 2 posts)
- Episode 14: Anthropic slowdown reflects market share loss to OpenAI, not AI fatigue(2026-11-10, 2 posts)
- Episode 15: Researchers complain Anthropic's Refusals Break Automated AI Research Loops(2026-11-10, 2 posts)
- Episode 16: Anthropic system card reveals Fable/Mythos 5.2 capabilities deliberately capped on AI R&D tasks(2026-11-10, 10 posts)
- Episode 17: Debate Over METR Speedup: Three Open Questions on Reality, Cause, and Compounding(2026-11-10, 3 posts)
- Episode 18: Safety researcher updates ASI timeline: ~25% odds by end of 2027(2026-11-10, 6 posts)
Primary sources
- Independent assessment suggests Anthropic's AI research is accelerating—with AI's help — Aiden_Tech_Ai · 2026-11-08
- [source] METR estimates Anthropic's benchmark progress accelerated 1.9x, with AI likely driving the gain — thione · 2026-11-10
- METR's report on Anthropic's RSI progress was redacted as "commercially sensitive"; call to subpoena — trevposts · 2026-11-10
- [source] METR publishes report assessing whether AI is accelerating AI R&D itself — itsOmSarraf_ · 2026-11-10
- [source] METR's independent assessment finds ~2x AI R&D speedup at Anthropic, heavily redacted — austinc3301 · 2026-11-10
- METR report: AI R&D at Anthropic sped up ~2x, sparking AI timeline debate — trevposts · 2026-11-10
- METR report: AI R&D at Anthropic has sped up ~2x as AI starts accelerating AI research — iruletheworldmo · 2026-11-10
- METR insider distills takeaways from its AI R&D acceleration assessment of Anthropic — sebkrier · 2026-11-10
- METR publishes independent Anthropic AI R&D assessment with most key evidence redacted — lxrjl · 2026-11-10