🔍 Read the full analysis: AI Research And AI Assistance: METR’s Findings From Anthropic on ThorstenMeyerAI.com
Get hardware and tech essentials delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
METR’s independent assessment estimates that Anthropic’s progress on well-specified AI capability tasks accelerated about 1.9 times after October 2025, with AI assistance likely accounting for much of the increase. The estimate relies on internal data, and Anthropic redacted a majority of the evidence METR considered most informative, limiting public scrutiny.
METR, a nonprofit that evaluates frontier AI systems, said in an assessment published November 5 that Anthropic’s progress on measured AI capabilities accelerated about 1.9 times after October 2025. The evaluator attributes most of the increase, with uncertainty, to AI helping researchers develop AI systems; it also says Anthropic redacted a majority of the evidence METR found most informative, leaving important parts of the analysis difficult for outsiders to check.
METR’s central estimate compares Anthropic’s pace of capability progress before and after a statistical break point it places at October 7, 2025. Using a measure based on benchmark scores, the assessment estimates progress of roughly 15 points per year before the break and 29 points per year afterward. METR describes that change as 1.9 times the earlier pace, with a 95% confidence interval of 1.5 to 2.8 times. It says it would not be surprised by estimates ranging from 1.3 to 3 times.
The assessment draws on internal information rather than testing only publicly released models. METR staff reviewed benchmark results for unreleased models, training metrics, surveys and about 15 interviews with Anthropic staff. The work involved about 2.5 full-time-equivalent staff over 11 weeks, on Anthropic-provided devices. METR says it received no payment; Anthropic supplied laptops and free tokens and retained the right to redact non-public material.
METR estimates that perhaps about 80% of the acceleration may be attributable to AI use, while saying it is confident the change did not result from faster growth in training compute. That percentage is an estimate, not a directly measured share. Separately, Anthropic’s own measurement says Claude led 26% of its AI research and development work in August 2026, up from under 1% in February. The company also reported about 30,000 agents operating at a time on its main internal platform. Those figures describe Anthropic’s reporting, not an independent measurement by METR.
Infographic 01 · How fast is AI speeding up AI?
About twice the old pace, most likely because of AI
An outside evaluator's estimate from internal Anthropic data, plus Anthropic's own measurements. Wide uncertainty, much of the evidence redacted.
Acceleration of Anthropic’s benchmark progress (AECI) since October 2025, 95% confidence interval 1.5x to 2.8x. METR’s overall judgment: around twice the old rate; it “would not be surprised” by 1.3x or 3x.
Capability points per year
AECI trend of Anthropic models, break point 7 Oct 2025
All labs before October 2025: +17 a year, so 1.4x. Different task-difficulty assumptions: 1.1x to 2.9x.
Sources: METR assessment, 5 Nov 2026 · METR risk report review · Anthropic Institute measurements.
Powered by Thorsten Meyer AI
What Faster AI Research Could Mean
The findings bear on whether AI tools are already helping labs produce more capable AI systems faster. If that contribution grows, it could affect how quickly companies and regulators need to evaluate new capabilities and put safeguards in place. METR’s result is evidence about one company’s development process, not proof that AI systems can independently build their successors or that the same rate applies across the industry.
The distinction matters because Anthropic’s Responsible Scaling Policy defines a twofold pace of progress, plausibly attributable to automated research or engineering, as one trigger for safeguards related to automated research and development. METR says it is plausible the threshold may have been crossed by July 15, 2026. Anthropic’s November 2 SEC filing said the company “may imminently meet” its definition of dramatically accelerated progress. These statements indicate a possible warning threshold, not a finding that fully autonomous AI research is already happening or that a safety event has occurred.
For readers, the report’s limits are part of the news. The estimate concerns a consequential area of AI development, but the redactions mean outside researchers cannot inspect much of the material behind the conclusion. The result is informative, yet not fully independently reproducible from the public record.
Products Worth Considering
LG UltraGear 45GX900A-B 45-Inch Curved OLED Gaming Monitor
Apple MacBook Pro 16-inch with M1 Pro, 10-core CPU, 16-core GPU, 16GB RAM, 1TB SSD, Silver
Trezor Safe 7 Crypto Hardware Wallet
As an affiliate, we earn on qualifying purchases.
How METR Built Its Estimate
METR assessed Anthropic’s internal research and development between August 12 and October 29, 2026. Its main metric, AECI, adapts the Epoch Capabilities Index by combining scores across multiple benchmarks into a capability measure. The comparison is about the pace at which scores improved on specified tasks; it is not a direct measure of researcher productivity or the total time required to develop and deploy a model.
METR also compared frontier models across developers. On that broader series, it estimated a smaller acceleration: progress rose from about 17 to 24 points per year, or 1.4 times. The report offers two possible explanations for the difference: Anthropic may have been behind the frontier in 2025, making its later increase look larger, or its earlier systems may have been less tuned to the benchmarks, making the increase look smaller. METR says changing assumptions about task difficulty moves its Anthropic estimate from 1.1 to 2.9 times, although many other checks remain near twofold.
Anthropic’s policy, version 3.4, took effect on July 8, 2026. It identifies two potential automated-R&D triggers: an AI system able to replace the company’s research staff at competitive cost, or a doubling of the pace of progress plausibly attributable to automated research or engineering. The policy describes the threshold as an early warning. Anthropic has said it is not yet at recursive self-improvement, which it defines as an AI system fully autonomously designing and developing its successor.
““roughly two years of historical progress over the last year.””
— METR assessment
Evidence Limits and Open Questions
The public report does not provide all the material needed to independently test METR’s conclusion. METR says Anthropic redacted a majority of the evidence it found most informative. The company also did not complete a planned check of how it divides compute between safety work and capability development. The assessment’s conclusions therefore rest partly on information that readers and outside researchers cannot review.
The 1.9-times figure is an estimate sensitive to the benchmark set, the choice of comparison period and assumptions about task difficulty. METR’s cross-company estimate is lower, at 1.4 times, and the report describes multiple possible interpretations of that gap. The estimate that AI accounts for about 80% of the acceleration is also a judgment, not a directly observed causal breakdown. It remains unclear from the public material how much of the change reflects AI assistance, other changes in research practice, or measurement choices.
Nor does the assessment establish that Claude’s reported share of work translates directly into faster development across every stage, or that current systems can independently set research goals and build successor models. The public information does not settle whether Anthropic has crossed its policy threshold; METR calls that plausible, while the company’s filing says it may imminently meet its definition.
Policy Thresholds and Further Evidence
The next key development is whether Anthropic makes further determinations under its Responsible Scaling Policy and reports what safeguards follow if it concludes a trigger has been met. Its November 2 filing signals that the question is live, but the material provided here does not state that the threshold has been formally crossed or specify a new safeguard decision.
Further public evidence could clarify the estimate, including additional benchmark results, a fuller account of the redacted findings, and more information about how Anthropic measures AI contributions to research. METR’s assessment is a snapshot based on work completed through October 29, 2026; changes in models, methods or disclosure could affect future estimates. Until more details become available, the reported acceleration should be read as an independent but partly uncheckable estimate, not a definitive measure of the entire AI industry’s development pace.
Key Questions
What did METR estimate about Anthropic’s AI progress?
METR estimated that Anthropic’s progress on its measured capability tasks accelerated about 1.9 times after a break point dated October 7, 2025. Its 95% confidence interval is 1.5 to 2.8 times.
Does the report say AI is building AI without human involvement?
No. The assessment concerns AI assistance with research and development. Anthropic says it is not at fully autonomous recursive self-improvement, in which an AI system designs and develops its own successor.
How much of the acceleration did METR attribute to AI?
METR said perhaps about 80% may be attributable to AI use. This is a qualified estimate, not a directly measured causal share. METR also said it was confident the acceleration did not come from faster growth in training compute.
Why can’t readers fully verify the findings?
METR said Anthropic redacted a majority of the evidence it considered most informative. Anthropic supplied internal data and retained the right to redact non-public material, so much of the evidence behind the assessment is not available to outside researchers.
Has Anthropic confirmed it crossed its automated-R&D threshold?
The available source material does not say the threshold has been formally crossed. METR says crossing it by July 15 is plausible, and Anthropic’s November 2 filing said it “may imminently meet” its definition of dramatically accelerated progress.
Source: ThorstenMeyerAI.com
