🔍 Read the full analysis: Could AI Be Accelerating Its Own Progress? A Look At METR’s Findings 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 estimates that Anthropic’s progress on measured AI capabilities accelerated to about 1.9 times its earlier pace, with a 95% confidence interval of 1.5x to 2.8x. The nonprofit believes AI use accounts for much of the increase, but Anthropic redacted most 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 measured AI capabilities is now about 1.9 times its pace before October 2025. The assessment, based on 11 weeks of access to internal company data, suggests AI tools may account for much of the acceleration; however, METR said Anthropic redacted a majority of the evidence it found most informative, leaving important parts of its conclusion difficult for outsiders to verify.
METR’s central estimate compares Anthropic’s measured capability progress before and after a change point it places at October 7, 2025. Its measure, AECI, combines results from multiple benchmarks into a capability index. METR estimates the index rose by about 15 points per year before the change point and about 29 points per year afterward. The resulting 1.9x estimate has a 95% confidence interval of 1.5x to 2.8x. METR said it would not be surprised by an acceleration between 1.3x and 3x.
The estimate is not a claim that every part of AI development happens twice as quickly, or that researchers are twice as productive. METR defines a 2x acceleration as making, in one year, the amount of progress that would previously have taken two years. The report says roughly 80% of the current acceleration may be due to AI use, while METR is confident it did not result from faster growth in training compute. That share is an estimate, not a direct measurement of how much work AI performs.
METR worked with unreleased-model benchmark results, training metrics, surveys and about 15 staff interviews. The nonprofit said about 2.5 full-time-equivalent staff conducted the work on Anthropic devices and that it received no payment. Anthropic provided laptops and free tokens and retained the right to redact non-public material. METR said those redactions removed most of the evidence it considered most informative, and that it dropped a planned examination of how Anthropic allocates compute between safety work and capability development.
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 a Faster R&D Cycle Matters
The report matters because using AI to assist AI research could create a feedback loop: tools help researchers build more capable systems, which may then help with further development. METR’s findings offer an outside estimate of the pace of that process at one major AI company, rather than a forecast of when fully autonomous AI research will arrive. The estimate may inform discussions about whether existing safety procedures can keep pace with development.
METR’s results are also relevant to Anthropic’s own safety policy. The company’s Responsible Scaling Policy sets a trigger for automated research and development, including a threshold based on a doubling of the progress rate when the increase is plausibly attributable to research or engineering automation. METR says it is plausible that the threshold had been crossed by July 15, 2026. That is the evaluator’s assessment, not a confirmed determination that Anthropic’s policy was triggered.
The limits on disclosure matter alongside the headline number. Because outsiders cannot inspect much of the evidence, they cannot independently assess all the assumptions, measurements or alternative explanations behind the estimate. The report therefore provides a significant external assessment, but not a fully reproducible public record.
Products Worth Considering
Apple MacBook Air 13.6-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, Sky Blue
Apple MacBook Air 15-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, and Liquid Retina Display (Midnight)
Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite
As an affiliate, we earn on qualifying purchases.
The Measures Behind the Estimate
METR’s assessment differs from a typical outside evaluation of a finished, publicly available model. Its staff examined internal company information, including benchmark scores for models that had not been released. The work took place from August 12 through October 29, 2026, with Anthropic controlling access to non-public material and retaining the right to redact it.
METR’s estimate is higher than the 1.7x figure in Anthropic’s Haiku 5.5 system card. METR said one reason is that its analysis included newer, unreleased models. When it looked at frontier models across developers, rather than Anthropic models alone, it estimated a smaller acceleration: 1.4x, from 17 to 24 index points per year. METR said this difference could mean Anthropic had been behind the frontier in 2025, making its rebound look larger, or that its earlier models were less tuned to the benchmarks.
Anthropic’s Responsible Scaling Policy, version 3.4, took effect July 8, 2026. It defines automated R&D triggers that include a model capable of replacing the company’s research staff at competitive cost, or a doubling of the pace of progress plausibly attributable to automation. Anthropic’s risk report describes the threshold as a potential early warning, not proof that a particular danger has already arrived. In a November 2 SEC filing, the company said it “may imminently meet” its definition of dramatically accelerated progress.
““Perhaps ≈80%” of the current acceleration is due to AI.”
— METR
What the Redactions Leave Unknown
METR’s estimate remains subject to substantial uncertainty. Under different assumptions about task difficulty, the acceleration estimate ranges from 1.1x to 2.9x, although the report says most other checks remain near 2x. The comparison across all developers also produces a lower figure than the Anthropic-only analysis. Those results make the estimate sensitive to which models and benchmarks are included and how progress is measured.
Public readers cannot fully evaluate the underlying evidence because Anthropic redacted a majority of the material METR found most informative. The planned review of the company’s compute allocation between safety and capabilities was also dropped. It is not clear from the available material how those omissions affect the report’s conclusions or whether additional evidence will be made public.
The findings concern measured progress at Anthropic; they do not establish that the entire AI industry is accelerating at the same rate or that fully autonomous recursive self-improvement has begun. METR’s view that AI accounts for about 80% of the acceleration is explicitly tentative. The exact contributions of AI assistance, model tuning, benchmark changes and other factors remain uncertain.
Further Evidence and Policy Signals
The next questions are whether Anthropic will disclose more of the material behind METR’s assessment and whether the company will make a formal determination under its Responsible Scaling Policy. Anthropic’s November 2 filing said the company may imminently meet its definition of dramatically accelerated progress, while METR judged that crossing the policy’s early-warning threshold by July 15 was plausible. Neither statement, on its own, confirms that the threshold has been formally met.
Further assessments could test whether the reported acceleration persists, how much it varies across types of research tasks and whether similar patterns appear at other AI developers. Public comparisons will also depend on access to comparable data, including results from unreleased systems. Until that information is available, METR’s report is an important but limited snapshot of one company’s internal development pace.
Key Questions
What did METR estimate about Anthropic’s AI progress?
METR estimated that Anthropic’s progress on its measured capability index accelerated by about 1.9x after a change point it places at October 7, 2025. Its 95% confidence interval is 1.5x to 2.8x.
Does the report say AI is building AI by itself?
No. The report describes AI assisting people with AI research and development. Anthropic says it is not yet at the stage where an AI system fully autonomously designs and develops its successor.
How much of the acceleration does METR attribute to AI?
METR estimates that perhaps about 80% of the acceleration is due to AI use. The report presents this as a judgment, not a directly measured share.
Why is the estimate difficult to check independently?
Anthropic redacted a majority of the evidence METR said it found most informative. METR also dropped a planned check of how the company splits compute between safety and capability work.
Does METR say Anthropic has crossed its safety-policy threshold?
No. METR said it was plausible the threshold had been crossed by July 15, 2026. Anthropic’s November 2 filing said it may imminently meet its definition of dramatically accelerated progress; neither statement confirms a formal threshold determination.
Source: ThorstenMeyerAI.com
