Share of Anthropic's observed AI-capability acceleration in October 2026 attributable to AI assistance (percent)

The only direct estimate comes from METR's November 2026 independent assessment, which, after a high-access evaluation at Anthropic, guesses that a majority, perhaps about 80%, of the October 2026 acceleration is due to AI, and calls that judgment highly uncertain. METR reports that capability index growth since October 2025 ran at about 1.9 times the earlier rate (95% interval 1.5 to 2.8), that the acceleration was not due to faster scaling of training compute, and that staff interviews and surveys are consistent with AI use roughly doubling the pace of work, while noting that self-reports are hard to interpret and that non-AI factors played a large role in late 2025.

The graph's estimate sits below METR's figure, with a median near 65% and a 90% interval of roughly 25% to 95%. Three considerations pull it down and widen it. First, Anthropic redacted most of the evidence METR found most informative, so the attribution rests largely on evidence not shown, assessed by an examiner whose access Anthropic granted and whose report Anthropic could redact. Second, METR itself lists as a limitation that non-AI factors (hiring, research direction, organizational changes, an unusually slow 2023 to 2025 baseline) could explain most or all of the acceleration, and attribution of a growth-rate change among co-occurring causes is intrinsically weakly identified. Third, the interview evidence METR summarizes is mixed (two yes, three leaning yes, four unclear, two no on a 2x speedup). In favor of a large share: compute scaling is ruled out as the driver, AI use is pervasive in Anthropic's R&D, and METR judged no other investigated factor able to explain sustained acceleration. A small share of probability is held outside any value, for the case in which the acceleration itself is not real.