🔍 Read the full analysis: METR On AI Development: What Anthropic Revealed And Redacted 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’s November 5 assessment estimates that Anthropic’s progress on measured AI capabilities accelerated 1.9 times after October 2025, with a 95% confidence interval of 1.5x to 2.8x. METR attributes much of the change to AI use in research and development, but Anthropic redacted most of the evidence METR considered most informative, limiting public scrutiny.
METR, a nonprofit that evaluates frontier AI systems, estimates that Anthropic’s measured AI capabilities have been improving at about 1.9 times their earlier pace since October 2025, based on an assessment published November 5. The report says AI assistance in research and development likely accounts for much of the acceleration, but also says Anthropic redacted a majority of the evidence METR found most informative, leaving the public with limited ability to examine the basis for its conclusions.
METR staff spent 11 weeks working with Anthropic’s internal data, including benchmark results for unreleased models, training metrics, staff surveys and about 15 interviews. The work ran from August 12 to October 29, 2026, with about 2.5 full-time-equivalent staff. METR says it received no payment; Anthropic supplied laptops and free tokens and retained the right to redact non-public information.
METR’s main estimate compares Anthropic’s capability trend before and after a statistical break point it places at October 7, 2025. Its index rose by about 15 points per year before that point and about 29 points per year afterward, producing the 1.9x estimate. METR gives a 95% confidence interval of 1.5x to 2.8x and says it would not be surprised by a result between 1.3x and 3x.
The estimate is not a direct measure of how quickly all AI research is being completed. METR defines a 2x acceleration as making two years of historical progress in one year, based on measured capabilities. The report says its estimate is higher than the 1.7x figure in Anthropic’s Haiku 5.5 system card, in part because METR included newer models that were not released publicly.
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 researchers build more capable systems, they could shorten the time between successive advances. That possibility matters for AI safety oversight: policies and evaluations may need to keep pace with a development cycle that is changing faster than before. METR’s result is an estimate of capability progress, not proof that AI systems are independently designing their successors.
METR attributes perhaps about 80% of the acceleration to AI, while saying it is confident the change did not result from faster growth in training compute. That attribution is METR’s interpretation of the data, not an independently verified causal measurement available to readers. The report’s redactions make it harder for outside researchers to assess that interpretation and the underlying evidence.
The finding also has practical relevance because Anthropic’s own safety policy includes a threshold for dramatically accelerated progress. METR says it is plausible the company may have crossed its early-warning line by July 15, 2026. Whether the policy threshold was formally triggered, and what action that would require, is not established by the supplied report material.
Products Worth Considering
Logitech MX Keys Combo for Business Keyboard and Mouse – Graphite
Sonos Era 100 Wireless Smart Speaker, Black
DJI Mini 5 Pro Fly More Combo with DJI RC 2
As an affiliate, we earn on qualifying purchases.
How METR Measured Anthropic’s Progress
METR used AECI, Anthropic’s version of the Epoch Capabilities Index, which combines scores across benchmarks into an overall capability measure. The organization’s analysis focused on Anthropic rather than the entire AI industry. When METR instead examined frontier models across developers, it found a smaller increase: from about 17 to 24 index points per year, or 1.4x.
METR offers more than one possible explanation for the difference. Anthropic may have been behind the frontier in 2025, which could make its subsequent acceleration look larger; alternatively, earlier Anthropic models may have been less tuned to the benchmarks, which could make the increase look smaller. Under different assumptions about task difficulty, the estimate ranges from 1.1x to 2.9x, although METR says most other checks remain near 2x.
Anthropic’s Responsible Scaling Policy version 3.4, effective July 8, 2026, identifies two possible triggers related to automated research and development: a model able to replace the company’s research staff at competitive cost, or a doubling in the pace of progress plausibly attributable in substantial part to automating research or engineering. Anthropic describes this threshold as an early warning, not proof that a danger has already materialized.
““A 2x acceleration means that 2 years of historical progress is made in 1 year.””
— METR
Redactions Limit Public Verification
The report says Anthropic redacted a majority of the evidence METR found most informative. Anthropic also dropped a planned check of how compute is divided between safety work and capability development. The public account therefore does not provide a complete basis for independently reproducing or evaluating the assessment.
The estimate depends on the benchmark index, the models included and assumptions about task difficulty. METR’s alternate industry-wide measure is lower, and its sensitivity analysis spans 1.1x to 2.9x. The supplied material also does not spell out the redacted evidence or quantify precisely how much the redactions affect confidence in each finding.
Anthropic reported that Claude led 26% of its AI research and development work in August 2026, compared with less than 1% in February, with about 30,000 agents operating at a time on its main internal platform. These are Anthropic’s own measurements; the material does not define “leads” in enough detail for readers to interpret it as a measure of autonomous research or completed work.
Policy Threshold and Further Evidence
The next points to watch are whether Anthropic clarifies the status of its Responsible Scaling Policy threshold and whether additional evidence becomes public. An SEC filing dated November 2 said Anthropic “may imminently meet” its definition of dramatically accelerated progress. METR’s assessment says crossing the threshold by July 15 is plausible, but the supplied material does not confirm that Anthropic formally declared it met the trigger.
Further assessments could test whether the measured acceleration persists and whether it appears across other developers, benchmarks and types of research work. Until more of the underlying data or methodology is available, the 1.9x figure should be read as METR’s estimate, with meaningful uncertainty around its size and cause.
Key Questions
What did METR estimate about Anthropic’s AI progress?
METR estimated that Anthropic’s measured capability progress accelerated 1.9x after a break point it places in October 2025. Its 95% confidence interval is 1.5x to 2.8x.
Does the report show that AI is autonomously building its own successor?
No. The report describes AI helping people with AI research and development. Anthropic’s definition of recursive self-improvement involves a system autonomously designing and developing its successor; Anthropic says that has not yet happened.
How much of the acceleration did METR attribute to AI?
METR estimated that perhaps about 80% of the acceleration came from AI use. This is an attributed estimate, not a proven causal measurement.
Why are the findings difficult to verify independently?
METR says Anthropic redacted a majority of the evidence it found most informative. The company also dropped a planned analysis of how it divides compute between safety and capability work.
Did Anthropic confirm that it crossed its policy threshold?
The supplied material does not confirm a formal threshold declaration. An SEC filing said Anthropic “may imminently meet” its definition, while METR said crossing it by July 15 was plausible.
Source: ThorstenMeyerAI.com
