---
id: "2026-11-05-metr-review-anthropic-automated-rd-acceleration"
url: "https://postcutoff.com/e/2026-11-05-metr-review-anthropic-automated-rd-acceleration/"
as_of: "2026-11-09T21:13:00+01:00"
date: "2026-11-05"
date_precision: day
category: policy-safety
importance: 5
confidence: high
status: [Confirmed]
sources: 9
editor: human editor
human_review: null
version: "2026-11-09"
---

As of: 2026-11-09 21:13 CET.
Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/
Canonical page: https://postcutoff.com/e/2026-11-05-metr-review-anthropic-automated-rd-acceleration/

# METR review of Anthropic's risk report

Full title: METR review of Anthropic's risk report: AI R&D at Anthropic has sped up about 2x, and the 2x acceleration threshold was 'plausibly' crossed by July 15

On Nov 5, 2026 METR published its review of the 'Risks from automated R&D' section of Anthropic's August 2026 Risk Report, made with access to nonpublic Anthropic data: its leading indicators show Anthropic's AI progress accelerating 1.3x–2.8x (Anthropic's private data imply ~1.97x), METR says it is plausible Anthropic had crossed its own '2x acceleration' threshold by July 15, and Anthropic redacted most of the evidence METR found most informative from the public version.

## Key facts
- The Verge (Nov 7) reported researchers' alarm at the METR review's findings on Anthropic automating parts of AI development
- Nov 7: commentator @kimmonismus (~210k views) said the independent AI 2027 tracker (ai2027-tracker.com, not affiliated with AI Futures Project) puts some Anthropic internal models ahead of the AI 2027 curve on METR's AECI figures; on Nov 8 the tracker's public page showed no AECI comparison, so this is unverified
- Ajeya Cotra (Planned Obsolescence, Nov 6; HN 324 points): argues the 'dramatic acceleration' threshold has already been crossed; METR attributes ~80% of the ~2x speed-up to AI (~1.8x vs a no-AI counterfactual); she estimates ~10x labour uplift (staff self-report ~12x, undocumented) and says smooth compounding would imply superintelligence within about a year
- Second wave Nov 5-7: Reuters ('improving nearly twice as fast with help from AI'), AP, Business Insider and ABC News covered the report; a LessWrong shortform doubting that AI R&D is 'ingenious' reached 203 HN points; Jeffrey Ladish noted the best evidence was redacted (BigGo)
- METR worked with internal Anthropic information since August 2026; full report ~100 pages, submitted to Anthropic
- Leading-indicator acceleration: 1.3x–2.8x depending on definition and period; METR's preferred public wording 'around 2'
- Anthropic's private report implies ~1.97x; METR calls the public wording 'less than a factor of 2' misleading
- 2x Acceleration Threshold: Anthropic says not crossed; METR: 'we think it is plausible that it had been' by July 15, though most researchers at both organisations lean 'not crossed'
- Full Substitution Threshold: not crossed (METR finds Anthropic's argument broadly compelling, with less confidence than before)
- Both agree catastrophic risk from July 15th models was low; METR cannot rule out substantial further acceleration within ~6 months, which could raise risk to 'high'
- AECI estimate rose from ~1.5x to ~1.9x with extra benchmark data, per METR
- Redaction: 'Anthropic redacted a majority of the evidence we found most informative… the public must rely on our judgment more than we consider ideal' (METR on X, 287k views); launch thread 1.0M views
- Anthropic co-founder Sam McCandlish (who leads much of its safety work) welcomed the report as 'an independent, external view' (X, 458k views)
- The Information's summary: Anthropic's models improving about twice as fast as between March 2024 and October 2025, with internal AI use driving most of the acceleration

## What happened

METR, given employee-level access to nonpublic Anthropic information from August 2026, reviewed the automated-R&D section of Anthropic's August 2026 Risk Report. It agrees catastrophic risk from the July 15 models was low, but disagrees with how the public report describes the speed-up: by its leading indicators AI progress at Anthropic is roughly twice as fast as before, and it is "plausible" that Anthropic's own 2x acceleration threshold had already been crossed. METR asks Anthropic to revise the public wording and to state the 1.3x–2.8x range.

## Why it matters

It is the first external, data-backed estimate that a frontier lab's own AI is measurably speeding up its AI research, the mechanism behind "recursive self-improvement" worries, and also a test of how much an embedded evaluator can tell the public when the lab controls redactions.

## Your AI and this story

- GPT-6 Astra (training cutoff April 2026): 189 days after its cutoff
- Claude Opus 5.5 (training cutoff June 2026): 128 days after its cutoff
- Gemini 4 Argon (training cutoff March 2026): 219 days after its cutoff
- Grok 4.7 (training cutoff May 2026): 158 days after its cutoff

## Sources

1. [METR: review of the R&D section of Anthropic's August 2026 Risk Report](https://metr.org/blog/2026-11-05-rd-section-anthropic-risk-report-august-2026-review/) (metr.org, official)
2. [The Verge: Why a new report on Anthropic automating AI development is alarming researchers](https://www.theverge.com/ai-artificial-intelligence/1029126/anthropic-metr-report-accelerating-ai-development) (theverge.com, press)
3. [The Information: Anthropic's R&D speed may have doubled due to AI, researchers find](https://www.theinformation.com/briefings/anthropics-r-d-speed-may-doubled-due-ai-researchers-find) (theinformation.com, press)
4. [kimmonismus on X: AI 2027 tracker update](https://x.com/kimmonismus/status/2118975993545105593) (x.com, discussion)
5. [METR on X: launch thread](https://x.com/METR_Evals/status/2118412183860351084) (x.com, discussion)
6. [METR on X: on Anthropic's redactions](https://x.com/metr_evals/status/2118412194702598339) (x.com, discussion)
7. [Sam McCandlish on X: response to METR's report](https://x.com/sam_mccandlish/status/2118422266669371519) (x.com, discussion)
8. [Ajeya Cotra: AI is dramatically accelerating AI research](https://www.planned-obsolescence.org/p/ai-is-dramatically-accelerating-ai) (planned-obsolescence.org, discussion)
9. [LessWrong: I struggle to believe that AI R&D is ingenious (shortform)](https://www.lesswrong.com/posts/DALNiKBjMGMF8grme/arjun-panickssery-s-shortform) (lesswrong.com, discussion)

## Changes

- 2026-11-09: Verge coverage (news agent)
- 2026-11-08: Added AI 2027 tracker claim (unverified)
- 2026-11-07: Cotra essay, wire coverage and sceptic thread (news agent)
- 2026-11-06 (filed): Created

## Related

- 2026-10-15: [Anthropic gives METR employee-like access and veto-free publication rights](https://postcutoff.com/e/2026-10-15-anthropic-metr-employee-level-access/index.md)
- 2026-08-01: [Anthropic publishes August 2026 Risk Report under its RSP](https://postcutoff.com/e/2026-08-01-anthropic-risk-report-august-2026/index.md)
- People: [Ajeya Cotra](https://postcutoff.com/person/ajeya-cotra/), [Sam McCandlish](https://postcutoff.com/person/sam-mccandlish/)
- Archived: 4 posts, listed in https://postcutoff.com/e/2026-11-05-metr-review-anthropic-automated-rd-acceleration/index.json