🔍 Read the full analysis: METR Looks Inside Anthropic At AI’s Role In AI Development on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
METR’s November 5 assessment estimates that Anthropic’s progress on measured AI capabilities accelerated about 1.9 times after October 2025. The nonprofit believes AI use likely explains much of the change, but says Anthropic redacted most of the evidence it found most informative, limiting public verification.
METR, a nonprofit that evaluates frontier AI systems, reported on November 5 that Anthropic’s measured model progress accelerated about 1.9 times after October 2025. The assessment, based on 11 weeks of access to company data, says AI use likely accounts for most of the increase, while warning that Anthropic redacted a majority of the evidence METR found most informative.
METR’s main estimate compares Anthropic’s progress on its version of the Epoch Capabilities Index, or AECI, before and after a statistical break point it places at October 7, 2025. The measured pace rose from about 15 index points per year to 29, yielding the 1.9x estimate. METR describes that as roughly two years of historical progress occurring in one year. Its stated 95% confidence interval is 1.5x to 2.8x; the organization says it would not be surprised by a result between 1.3x and 3x.
The estimate is specific to the models and benchmark measure used, not a finding that every part of Anthropic’s research or product development is twice as fast. METR says its result is higher than the 1.7x figure in Anthropic’s Haiku 5.5 system card, largely because METR included newer models that were not released. When it compared frontier models across developers rather than Anthropic models alone, it estimated a smaller rise, from 17 to 24 points per year, or 1.4x.
METR’s work drew on benchmark scores for unreleased models, training metrics, surveys and about 15 staff interviews. The nonprofit says roughly 2.5 full-time-equivalent staff worked on the assessment using Anthropic devices from August 12 to October 29. METR says it received no payment; Anthropic supplied laptops and free model tokens 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
What a Faster R&D Pace Could Mean
The assessment addresses whether AI is beginning to speed up the development of later AI systems, a question with potential consequences for both technology progress and safety planning. If the pace of research and engineering rises because AI tools are helping researchers, labs may reach more capable systems sooner than earlier projections assumed. The report does not establish that fully autonomous AI research is happening; its account is of AI assisting people, who continue to set goals and make decisions.
Anthropic’s Responsible Scaling Policy sets a potential early-warning trigger for automated research and development. Its version 3.4, effective July 8, 2026, includes a threshold based on a doubling of progress relative to a baseline when the acceleration is plausibly substantially attributable to automated research or engineering. METR says it is plausible Anthropic had crossed that threshold by July 15. That is METR’s interpretation, not confirmation that Anthropic formally determined the trigger had been met.
The public evidence gap matters because the estimate could inform debate about whether a company’s safety safeguards should change as its development process speeds up. METR’s figures give outsiders a rare view of internal progress measures, but its account of extensive redactions makes independent scrutiny difficult. The report therefore offers both a measurement and a limit on how confidently readers can assess it.
Products Worth Considering
Kindle Paperwhite 16GB (2024) – 7 Glare-Free Display, Weeks Battery Life – Black
Apple MacBook Air 15-inch with M5 chip, 16GB memory, 512GB SSD, and Liquid Retina display (Midnight)
Apple MacBook Air 13-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, 12MP Center Stage Camera, Touch ID, Wi-Fi 7, Midnight
As an affiliate, we earn on qualifying purchases.
How METR Measured Anthropic’s Progress
Unlike an evaluation that tests only a publicly released model, METR worked with internal company data. Its AECI measure combines results across multiple capability benchmarks into one index. The organization estimated a change in the trend around October 2025, then compared the rate before and after that point. The acceleration definition concerns the pace of measured capability progress: a 2x rate means two years of baseline progress in one year, not that all research tasks or timelines have been cut in half.
The findings arrive amid Anthropic’s own disclosures about AI’s growing role in its work. According to the source report, Anthropic’s internal measurement says Claude led 26% of its AI research and development work in August 2026, compared with under 1% in February. The company also reported about 30,000 agents working at a time on its main internal platform. These are company-reported figures; the available material does not detail precisely how “leads” is defined.
Anthropic’s policy describes automated R&D as a risk area and says a capability that could replace its research staff at competitive cost is one possible trigger. The policy also describes a progress-doubling threshold. In a November 2 SEC filing, Anthropic wrote that it “may imminently meet” its definition of dramatically accelerated progress. The filing’s wording is prospective and does not say the threshold had already been reached.
““Perhaps ≈80%” of the current acceleration is due to AI.”
— METR
Redactions Limit Public Verification
METR says Anthropic redacted a majority of the evidence it found most informative. The source material does not specify the full content of those redactions, so readers cannot independently review all the underlying data or determine how the withheld evidence might affect the estimate.
The cause of the estimated acceleration is also uncertain. METR believes perhaps 80% is attributable to AI and says it is confident the increase did not come from faster growth in training compute. That attribution remains an estimate. METR also notes that the result changes with methodological assumptions: varying assumptions about task difficulty moves the estimate from 1.1x to 2.9x. Its cross-company comparison produces 1.4x rather than 1.9x, and METR offers two possible explanations involving Anthropic’s earlier position relative to the frontier and differences in benchmark tuning.
The supplied material also says METR dropped a planned check of how Anthropic divides compute between safety and capability work. It does not provide enough detail to assess what that analysis would have shown. The report’s findings should not be read as proof that recursive self-improvement has begun or that Anthropic’s policy threshold has definitively been crossed.
Policy Decisions and Further Evidence
The next key development is whether Anthropic determines that its Responsible Scaling Policy threshold has been met and what safeguards or disclosures would follow. METR’s assessment says the threshold may plausibly have been crossed by July 15, while Anthropic’s November 2 filing says it may be imminent. Neither statement, as presented in the source material, confirms a formal determination.
Further public detail about the redacted evidence, the benchmark data and the attribution of the acceleration would allow outside researchers to test the estimate more fully. The available source does not identify a date for additional disclosures or a follow-up evaluation. For now, the 1.9x figure is METR’s best estimate from an unusually detailed but partly concealed review of Anthropic’s internal work.
Key Questions
What did METR estimate?
METR estimated that Anthropic’s measured capability progress accelerated about 1.9x after a break point in October 2025, based on its AECI measure. The estimate has a 95% confidence interval of 1.5x to 2.8x.
Does the report show that AI is building AI without human involvement?
No. METR’s assessment concerns AI helping with research and engineering. The source describes humans as continuing to choose goals; it does not establish fully autonomous development of successor systems.
Why is the acceleration estimate uncertain?
METR says Anthropic redacted most of the evidence it found most informative. The estimate also shifts under different assumptions: METR reports a 1.1x to 2.9x range when it varies assumptions about task difficulty, and a 1.4x result in a comparison across developers.
Has Anthropic confirmed that its policy threshold was crossed?
The source material does not report such confirmation. METR says crossing the threshold by July 15 is plausible, while Anthropic’s November 2 filing says it may imminently meet its definition of dramatically accelerated progress.
Source: ThorstenMeyerAI.com
