🔍 Read the full analysis: Inside Anthropic: METR’s Findings On AI And Faster Development on ThorstenMeyerAI.com
Get smart everyday buys 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 estimates that Anthropic’s progress on measured AI capabilities accelerated to about 1.9 times its earlier rate, and says AI use likely accounts for most of the increase. The assessment draws on 11 weeks of access to internal data, but Anthropic redacted much of the evidence METR considered most informative.
METR, a nonprofit that evaluates frontier AI systems, said on November 5 that Anthropic’s progress on measured model capabilities is running at about 1.9 times its earlier pace, with AI use likely responsible for most of the acceleration. The assessment is based on 11 weeks of access to internal company data, but METR said Anthropic redacted a majority of the evidence it found most informative, limiting what outsiders can verify.
METR’s estimate compares Anthropic’s capability progress before and after a change point it places at October 7, 2025. Its measure, AECI, combines results across multiple benchmarks. METR estimates the score rose by about 15 points a year before that date and about 29 points a year afterward. The nonprofit describes the change as roughly two years of historical progress occurring in one year.
The estimate is not a direct measure of how much faster individual researchers work. METR says it concerns progress on well-specified tasks; other stages of AI development may still take time. Its reported 95% confidence interval for the acceleration is 1.5x to 2.8x, and METR said it would not be surprised by a result between 1.3x and 3x. Different assumptions about task difficulty move the estimate from 1.1x to 2.9x.
METR attributes perhaps 80% of the acceleration to AI use, while saying it is confident the increase did not result from faster growth in training compute. That share is an estimate, not a directly observed breakdown. The assessment used benchmark results from unreleased models, training metrics, surveys and about 15 staff interviews. METR said roughly 2.5 full-time-equivalent staff worked on the project using Anthropic devices; the company supplied laptops and free tokens, took no payment, and retained the right to redact non-public information.
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 Faster AI Research Matters
If AI tools help build more capable AI systems, development could speed up without a matching increase in the amount of computing hardware used for training. That matters for safety planning and oversight: a faster research cycle can leave less time to evaluate systems and put safeguards in place before capabilities change.
The result also bears on Anthropic’s own safety framework. Its Responsible Scaling Policy, version 3.4, identifies a doubling in the pace of progress plausibly attributable to automated research or engineering as one trigger for its “automated R&D” safeguards. METR said it was plausible that Anthropic’s early-warning threshold had been crossed by July 15, 2026. That is METR’s assessment, not a confirmed finding that Anthropic has met the policy trigger.
The findings are relevant beyond one company, but should not be read as a sector-wide measurement. When METR compared frontier models from all developers, it estimated a smaller acceleration of 1.4x, from about 17 to 24 index points per year. That difference shows how conclusions can depend on which systems and time periods a study includes.
Products Worth Considering
Apple MacBook Pro 14.2-Inch Laptop with M2 Pro, 16GB Memory, 1TB SSD, 19-Core GPU, Space Gray
Dell XPS 13 9350 Laptop, 13.4-Inch WUXGA Display, Intel Core Ultra 7 256V, 16GB RAM, 1TB SSD
Dell XPS 13 9345 Laptop, 13.4-Inch FHD+ 120Hz, Snapdragon X Elite, 32GB RAM, 1TB SSD
As an affiliate, we earn on qualifying purchases.
How METR Measured Anthropic’s Progress
METR’s work differs from the group’s usual external evaluations of finished AI models. For this assessment, its staff examined internal Anthropic data over 11 weeks, from August 12 to October 29, 2026. The source account says the researchers worked on Anthropic devices and reviewed benchmark scores, training metrics, surveys and interviews with staff.
Anthropic separately reported that, in August 2026, Claude led 26% of its AI research and development work, compared with less than 1% in February. The company also reported about 30,000 agents working at a time on its main internal platform. These are Anthropic’s own figures; “leads” does not mean the systems autonomously conduct all research or decide what goals to pursue.
Anthropic defines recursive self-improvement as an AI system fully autonomously designing and developing its successor. The company has said it is not at that point and that such a development is not inevitable. The METR report instead concerns AI assisting human research and engineering. In a November 2 SEC filing, Anthropic wrote that it “may imminently meet” its definition of dramatically accelerated progress.
““Perhaps ≈80%” of the current acceleration is due to AI.”
— METR
Limits of the Redacted Evidence
The public cannot fully inspect the evidence behind METR’s conclusion. The report’s account says Anthropic redacted a majority of the evidence METR found most informative. METR also dropped a planned examination of how Anthropic divides computing resources between safety work and capability development. The available material does not show precisely which findings were withheld or how those omissions affect each part of the analysis.
The 1.9x figure is an estimate shaped by the benchmark index, the selected models and the choice of baseline. METR says including unreleased Anthropic models makes its estimate higher than the 1.7x figure in Anthropic’s Haiku 5.5 system card. It also offers competing explanations for the difference between Anthropic-only and industry-wide results: Anthropic may have started behind the frontier in 2025, or its earlier models may have been less tuned to the benchmarks.
It remains unclear whether Anthropic has formally determined that its Responsible Scaling Policy threshold is met, what safeguards would follow if it is, and how much the reported acceleration translates into shorter end-to-end development timelines. METR’s findings do not establish that AI systems are independently directing or completing the creation of their successors.
Policy Thresholds and Further Evidence
The next concrete point to watch is whether Anthropic says it has reached the automated R&D trigger in its Responsible Scaling Policy, and what measures it would apply if so. METR’s assessment says crossing the threshold by July 15 was plausible, while Anthropic’s filing said it may imminently meet its definition; neither statement confirms a formal determination.
Further public detail could help researchers assess the estimate’s sensitivity to redactions, benchmark selection and the inclusion of unreleased models. The available source material does not give a date for additional disclosures or a planned follow-up assessment. Until more evidence is available, the 1.9x result remains an independent estimate with material limits on outside verification.
Key Questions
What did METR estimate about Anthropic’s AI progress?
METR estimated that Anthropic’s progress on its measured capabilities accelerated to about 1.9 times the pace it observed before October 2025. The estimate has a reported 95% confidence interval of 1.5x to 2.8x.
Does the report say AI is building AI without human involvement?
No. The report concerns AI assisting people with research and engineering. Anthropic defines recursive self-improvement as a system autonomously designing and developing its successor, and says it is not there yet.
How much of the acceleration does METR attribute to AI?
METR estimates that perhaps 80% of the acceleration is due to AI use. It is an estimate, not a directly measured share of all research work.
Why can’t readers fully verify the findings?
Anthropic redacted a majority of the evidence METR considered most informative and retained the right to withhold non-public information. METR also dropped a planned check of how computing resources are divided between safety and capability work.
Has Anthropic confirmed that it crossed its safety-policy threshold?
No formal confirmation is included in the source material. METR said crossing the threshold by July 15, 2026, was plausible; Anthropic said in a November 2 filing that it may imminently meet its definition.
Source: ThorstenMeyerAI.com
