🔍 Read the full analysis: What METR Found About AI’s Role In Developing AI At Anthropic 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 model capabilities have been advancing about 1.9 times as fast since October 2025 as during the preceding period. The nonprofit says AI use likely accounts for most of the acceleration, but Anthropic redacted much of the evidence METR found most informative, limiting outside scrutiny.
METR estimated that Anthropic’s AI capabilities are advancing about 1.9 times faster than they did before October 2025, in an independent assessment published November 5 that draws on 11 weeks of internal company data. The nonprofit said AI use likely accounts for most of the acceleration, while warning that Anthropic redacted a majority of the evidence METR considered most informative, leaving the public with limited ability to check the findings.
METR, a nonprofit that evaluates advanced AI systems, estimated that Anthropic’s capability progress rose from about 15 points per year to 29 points per year on its measure, the Anthropic-adjusted Epoch Capabilities Index. That produces an estimated acceleration of 1.9 times from a break point METR places around October 7, 2025. METR’s 95% confidence interval is 1.5 to 2.8 times; the organization said it would not be surprised by a result between 1.3 and 3 times.
The estimate is about the pace of measured capability gains, not a claim that every part of AI research now takes half as long. METR defines a 2x acceleration as achieving in one year as much progress as had historically taken two years. Its analysis used benchmark results, training metrics, staff surveys and about 15 interviews. Roughly 2.5 full-time-equivalent METR staff worked on the assessment from August 12 to October 29 using Anthropic devices. METR said it took no payment; Anthropic supplied laptops and free model tokens and retained the right to redact non-public information.
METR attributed perhaps about 80% of the acceleration to AI, presenting this as an estimate rather than a directly measured share. It said it was confident the increase did not result from faster growth in training compute. Anthropic’s own figures, as described in the assessment, say Claude led 26% of the company’s AI research and development work in August 2026, compared with less than 1% in February, and that about 30,000 agents were operating at a time on its main internal platform. Those figures describe Anthropic’s measures and should not be read as independent confirmation that AI caused the overall acceleration.
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
AI Tools and Anthropic’s Progress Rate
The report addresses whether AI is helping researchers build more capable AI systems faster. If that effect grows, it could shorten the time available to observe changes and adjust safety practices. METR’s result is relevant because it draws on internal data rather than relying only on public model releases, but it is an estimate of measured capability progress, not evidence that an AI system is autonomously designing its own successor.
The findings also intersect with Anthropic’s Responsible Scaling Policy. Its version 3.4, effective July 8, 2026, identifies a doubling in the rate of progress plausibly attributable to automated research or engineering as one trigger for safeguards. METR said it was plausible that Anthropic’s early-warning threshold had been crossed by July 15. Separately, an Anthropic filing with the U.S. Securities and Exchange Commission on November 2 said the company “may imminently meet” its definition of dramatically accelerated progress. That wording is a company statement, not a finding that the threshold has definitively been met.
For readers and policymakers, the assessment offers a substantial but qualified signal: AI may already be contributing to faster AI development at a major lab. The redactions matter because they make it harder for outsiders to evaluate the methods, assumptions and evidence behind a result with implications for safety policy.
Products Worth Considering
Samsung Galaxy Tab A11+ 11-Inch Tablet, 6GB RAM, 128GB Storage, Expandable, Gray
Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Black
Bose QuietComfort Headphones (2nd Gen), Wireless Over-Ear Noise-Cancelling Headphones, Black
As an affiliate, we earn on qualifying purchases.
How METR Measured Anthropic’s Gains
METR’s assessment differs from a conventional outside test of a released model. The evaluators worked with internal Anthropic material, including scores for unreleased models, training measures, staff surveys and interviews. Anthropic retained editorial control over what non-public material could be disclosed. METR said the company redacted a majority of the evidence the evaluators found most informative and that a planned examination of how Anthropic divides computing resources between safety work and capability development was dropped.
The central 1.9x figure applies to Anthropic’s measured progress. When METR compared frontier models from multiple developers, it estimated a smaller acceleration: 1.4 times, from about 17 to 24 points per year. METR offered two possible explanations for the difference: Anthropic may have been behind the frontier in 2025, making its later catch-up look unusually fast, or its earlier models may have been less tuned to the benchmarks. Different assumptions about task difficulty moved the estimate from 1.1 to 2.9 times, while METR said most other checks remained near 2x.
Anthropic’s policy uses “automated R&D” to cover more than a system independently creating a successor. The company describes full recursive self-improvement as an AI system autonomously designing and developing its own successor, and says current systems have not reached that point. In the present situation described by the report, AI assists human researchers, who continue to set goals and make decisions.
““Perhaps ≈80%” of the current acceleration is due to AI.”
— METR
Redactions Limit Public Verification
The 1.9x figure is an estimate, and its size depends in part on the benchmark measure, comparison period and assumptions about task difficulty. The broader frontier comparison produces a lower estimate of 1.4x, while alternative assumptions yield values from 1.1 to 2.9x. METR’s assessment does not settle which interpretation best captures Anthropic’s progress.
Public scrutiny is also constrained by Anthropic’s redactions. METR said most of the evidence it found most informative was withheld, and its planned analysis of computing allocations between safety and capability work was not completed. The source material does not provide enough detail to independently assess the underlying records or reproduce the full analysis. It is also unclear how much of the reported acceleration will persist, or how the company’s internal “leads” measure relates to the quality and impact of research tasks.
METR’s attribution that AI may account for roughly 80% of the change is a judgment based on its assessment, not proof of a precise causal share. The report does not establish that AI systems can independently conduct the full research and development process, nor does it show that recursive self-improvement has occurred.
Policy Thresholds and Further Evidence
The immediate next step is for Anthropic and outside evaluators to clarify how the reported results relate to the company’s Responsible Scaling Policy. Anthropic’s November 2 filing said it may imminently meet its definition of dramatically accelerated progress; the company’s public materials describe that threshold as an early warning, not proof that a particular danger has arrived.
Further disclosure could allow researchers to test METR’s assumptions and examine the evidence behind its estimate, including the relationship between AI-assisted work and observed benchmark gains. The assessment itself does not state when additional evidence or a follow-up evaluation will be published. Until more material becomes available, the reported acceleration and AI’s estimated contribution remain important but incompletely verifiable findings.
Key Questions
What did METR estimate about Anthropic’s AI progress?
METR estimated that Anthropic’s measured capability progress accelerated by 1.9 times after a break point around October 7, 2025, compared with the earlier rate. Its stated 95% confidence interval was 1.5 to 2.8 times.
Does the report say AI is building AI by itself?
No. METR’s account is that AI tools likely contribute substantially while people continue to conduct and direct research. Anthropic says its systems have not reached the point of autonomously designing and developing their own successors.
How much of the acceleration did METR attribute to AI?
METR estimated that perhaps about 80% came from AI use. That is an approximate judgment, not a precise causal measurement or a confirmed share.
Why is the report difficult to verify independently?
Anthropic redacted a majority of the evidence METR said it found most informative. A planned check of how the company divides computing resources between safety and capability work was also dropped, limiting what outsiders can assess from the public material.
Did METR conclude Anthropic had crossed its policy threshold?
Not conclusively. METR said crossing the threshold by July 15 was plausible. Anthropic’s November 2 filing said it “may imminently meet” its definition of dramatically accelerated progress; the available statements do not establish a definitive crossing.
Source: ThorstenMeyerAI.com
