🔍 Read the full analysis: What Anthropic’s Redacted Details Mean For METR’s AI Findings on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
METR estimates that Anthropic’s progress on selected AI capability measures accelerated to about 1.9 times its earlier pace, with much of the increase likely linked to AI-assisted research and development. The evaluation used internal data, but Anthropic redacted a majority of the evidence METR considered most informative, limiting public scrutiny.
METR, a nonprofit that evaluates frontier AI systems, estimated on November 5 that Anthropic’s progress on certain AI capability measures has accelerated to about 1.9 times its earlier rate. The assessment draws on 11 weeks of access to internal company data, but METR said Anthropic redacted a majority of the evidence it found most informative, limiting how fully outsiders can check the analysis.
METR compared Anthropic’s recent capability progress with its pace from March 2024 to October 2025. Its central estimate is a rise from about 15 points per year on Anthropic’s version of the Epoch Capabilities Index to about 29 points per year. METR places the estimated shift around October 7, 2025. The group’s reported estimate is 1.9x acceleration, with a 95% confidence interval of 1.5x to 2.8x; it said it would not be surprised by a result between 1.3x and 3x.
The report says the acceleration was likely driven largely by AI systems helping people conduct AI research and development, rather than by faster growth in training compute. METR suggested that about 80% of the acceleration may be attributable to AI, while describing that share as an estimate. Anthropic’s internal measurement, cited in the report, says Claude led 26% of the company’s AI R&D work in August 2026, compared with less than 1% in February. The company reported about 30,000 agents operating at a time on its main internal platform.
The estimate is not a direct measure of how quickly every part of AI development is moving. METR’s index combines benchmark results into a single capability measure, and the report’s findings concern Anthropic’s progress on tasks that can be specified and measured. When METR examined frontier models across developers, rather than Anthropic alone, the estimated pace increase was smaller: 1.4x, from about 17 to 24 points per year. The report said different assumptions about task difficulty could move Anthropic’s estimate from 1.1x to 2.9x.
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
Why the acceleration estimate matters
The report addresses whether AI is helping accelerate the development of more capable AI systems. If that contribution is substantial, labs and policymakers may have less time to test safeguards and respond to changing capabilities. METR’s estimate is relevant to that debate because it relies on internal measures and unreleased model results, rather than only public evaluations of finished systems.
But the reported acceleration is not evidence that AI systems are autonomously building their own successors. The source material describes a process in which AI helps human researchers and engineers, while people continue to set goals. METR’s estimate is also not a forecast of when a particular capability will arrive. It is a retrospective comparison of measured progress against an earlier baseline.
The redactions shape how much weight readers can place on the finding. METR had access to material unavailable to the public, but readers cannot independently inspect much of what the evaluators judged most useful. That leaves the report as a detailed outside assessment, not a fully reproducible public analysis.
Products Worth Considering
Apple MacBook Pro 14.2-Inch Laptop with M2 Pro, 16GB Memory, 1TB SSD, 19-Core GPU, Space Gray
Samsung 27-inch ViewFinity S7 4K UHD IPS Monitor
FLEXISPOT 36-Inch Standing Desk Converter, Black
As an affiliate, we earn on qualifying purchases.
How METR evaluated Anthropic
METR’s work differed from an outside evaluation that tests a released model. Its staff examined internal benchmark scores, training metrics and surveys, and interviewed about 15 Anthropic employees. Around 2.5 full-time-equivalent staff worked on the assessment from August 12 to October 29, 2026, using Anthropic devices. METR said it did not receive payment; Anthropic supplied laptops and free tokens and retained the right to redact non-public information.
The finding also intersects with Anthropic’s Responsible Scaling Policy. Version 3.4, effective July 8, 2026, identifies an accelerated-R&D threshold: progress in one year comparable to two years at the prior baseline, where the increase is plausibly attributable in substantial part to automated research or engineering. Anthropic’s risk report calls that threshold a potential early warning, not proof that a danger has already materialized.
In a November 2 SEC filing, Anthropic said it “may imminently meet” its definition of dramatically accelerated progress. METR said it was plausible the company’s early-warning threshold had already been crossed by July 15. These statements use related but not necessarily identical assessments; the source material does not establish that the filing itself confirms METR’s estimate.
““A 2x acceleration means that 2 years of historical progress is made in 1 year.””
— METR
What redactions leave unanswered
The public report does not disclose much of the evidence METR considered most informative. The source material says Anthropic redacted a majority of that evidence, but does not specify here which underlying data or analyses were removed. METR also dropped a planned check of how Anthropic divides compute between safety work and capability development. The reasons for each redaction and the effect of the omitted material on particular conclusions are not fully established in the available account.
There are also limits to the estimate itself. METR’s 1.9x figure depends on the selected benchmark index, the comparison period and assumptions about task difficulty. The report’s lower 1.4x estimate across frontier developers offers a different comparison, and the source material gives two possible explanations: Anthropic may have been behind the frontier in 2025, or its earlier models may have been less tuned to benchmarks.
It remains unclear how closely the estimate tracks progress across the full AI research process, including work not represented by benchmarks. The supplied material also does not provide enough detail to independently assess how the redacted evidence, the interview findings or the reported internal agent usage affected the final estimate.
Further evidence and policy thresholds
The immediate next step is greater public detail about the methodology and evidence, if Anthropic permits further disclosure or METR can publish additional checks. Readers will also be looking for updated measurements that show whether the estimated acceleration persists, changes or depends on particular models and benchmarks.
Anthropic’s policy threshold and its reporting on accelerated progress are likely to remain points of attention. The available source material does not specify a new METR assessment date or a further company disclosure. Until then, the 1.9x result should be read as METR’s estimate based on restricted internal evidence, with a stated range of uncertainty rather than a settled public measurement.
Key Questions
What did METR estimate?
METR estimated that Anthropic’s progress on its capability index accelerated by 1.9x relative to the earlier period from March 2024 to October 2025. Its 95% confidence interval was 1.5x to 2.8x.
Does the report say AI is developing AI autonomously?
No. The account describes AI assisting human researchers and engineers. Anthropic says it is not yet at the point of an AI system fully autonomously designing and developing its successor.
Why can’t the public fully verify the findings?
METR said Anthropic redacted a majority of the evidence it found most informative. The public account therefore does not expose all the data and analysis used in the assessment.
Is 1.9x the same as twice as much compute?
No. The figure compares the pace of measured capability progress with an earlier baseline. METR said it was confident the acceleration did not come from faster growth in training compute.
What is still uncertain?
The estimate varies with methodological assumptions, and the report’s redactions limit independent review. It is also unclear from the supplied material how closely the benchmark-based measure reflects progress across all AI research and development work.
Source: ThorstenMeyerAI.com
