AI-assisted research and development caused most of Anthropic's capability acceleration since October 2025.
Reassessed
Read the METR report directly; the earlier pass had relied on a summary. Corrected the instance passage to the report's attribution sentence. Added structure: the acceleration subclaim (assumes), the October 2026 attribution-share quantity (supports), and the non-AI late-2025 factors claim (contradicts), with for and against arguments. Placed the unplaced passage as a mention. Verdict unchanged at inconclusive with credence 0.5; confidence raised to 0.65 and marginal yield lowered to 0.3, because the decisive evidence is redacted. No dependents need notifying, since the verdict did not change.
Reassessed: still Inconclusive
verdict confidence 0.60 → 0.65 · credence 0.50
Updated claim instance
Instance ce0702d7-3316-46a7-ae9d-ca5e74372db0 (https://metr.org/blog/2026-11-05-anthropic-ai-rd-acceleration-assessment/): was stance=affirms, confidence=0.85; set stance="affirms", confidence=0.7, verbatimText="A majority of this acceleration today is likely attributable to Anthropic's extensive use of AI to do R&D, though there is significant uncertainty and other causes likely played a large role in acceleration in late 2025.", segmentId=null, speaker="METR (Cunningham, Rush, Whitfill, Wijk)", publication="METR Blog", sourceDate="2026-11-05", context="Executive summary of METR's high-access assessment. METR attributes a majority (perhaps about 80%) of the October 2026 acceleration to AI, but notes non-AI causes in late 2025 and expects cumulative attribution since October 2025 to be lower than the instantaneous figure; it affirms the claim for 'today' rather than the whole period.". The recorded passage was about autonomous contributions, not attribution of the acceleration. Having read the report, the executive-summary attribution sentence is the one that bears on this claim. Confidence lowered because METR affirms a majority share for the current (instantaneous) acceleration, while the claim covers the cumulative period.
Assessed
First pass, almost out of budget. Assessed as inconclusive at credence 0.5 from the summary of METR's evidence in the parent's seed note. No decomposition this pass. Marginal yield is high: the next pass should read the METR report and independent sources directly.
Assessed Inconclusive
verdict confidence 0.60 · credence 0.50
The claim rests almost entirely on a METR report on Anthropic, which gives a subjective estimate that roughly 80% of the acceleration as of October 2026 is due to AI assistance, with wide stated uncertainty. The same report says non-AI factors probably played a large role in late 2025, that faster scaling of training compute does not explain the change, that staff interviews centre on about a twofold speedup from AI tools, and that Anthropic staff themselves had no consensus on attribution. Anthropic redacted the detailed attribution estimates and most of the evidence behind them. The public therefore has to take the judgment on METR's word. Rival explanations have not been ruled out in public. They include catch-up from a growth rate that had been below the industry frontier, more compute for experiments as opposed to training, hiring and organizational changes, and benchmark fitting. The claim covers the whole period since October 2025, so the non-AI contribution METR acknowledges for late 2025 counts against "most" on a cumulative reading. Taken at face value, the evidence leans toward a large AI share by late 2026. Because the evidence is unpublished and the period includes late 2025, the cumulative "most" is roughly even odds. Publication of the redacted attribution analysis, or an independent decomposition of the speedup, would resolve the question.
Claim entered the graph