🔍 Read the full analysis: Is AI Making AI Development Faster? METR’s Anthropic Findings on ThorstenMeyerAI.com
Get the latest gadgets 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 AI capabilities advanced about 1.9 times faster after October 2025 than before, and says AI use likely accounts for most of the acceleration. The estimate draws on internal company data, but Anthropic redacted much of the evidence METR found most informative, leaving important details difficult for outsiders to verify.
METR, a nonprofit that evaluates frontier AI systems, estimated on November 5 that Anthropic’s AI capabilities have been improving at about 1.9 times their earlier pace, with AI assistance likely responsible for much 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 the evaluators considered most informative, limiting what the public can independently check.
METR compared Anthropic’s rate of progress on a combined measure of model capabilities before and after an estimated change point of October 7, 2025. It calculated that Anthropic’s scores rose by about 15 points per year before that date and about 29 points per year afterward. METR describes the resulting 1.9x estimate as roughly two years of historical progress in one year. Its 95% confidence interval is 1.5x to 2.8x; the organization said it would not be surprised by a result ranging from 1.3x to 3x.
METR said it believes AI use probably explains most of the acceleration, estimating that perhaps 80% is attributable to AI. It said faster growth in training compute did not account for the change. Those are the evaluators’ judgments, not a fully disclosed accounting of the sources of productivity. The report also notes that a comparison using frontier models from multiple developers produced a smaller increase: progress rose from about 17 to 24 points per year, or 1.4x.
The assessment drew on benchmark scores for unreleased models, training metrics, surveys and about 15 staff interviews. METR said roughly 2.5 full-time-equivalent staff worked on the project from August 12 to October 29, using Anthropic-provided laptops and free tokens, and that it took no payment. Anthropic retained the right to redact non-public material. The report says a planned analysis of how the company divides compute between safety and capabilities was dropped.
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 AI-Assisted R&D Matters
The report addresses whether AI is helping researchers develop more capable AI systems, a question with implications for how quickly companies and policymakers may need to respond to changing capabilities. If AI tools substantially shorten parts of research and engineering, a lab could make more progress in a given period without a comparable increase in computing resources. METR’s estimate suggests that this may already be happening at Anthropic, although the report does not establish that the pattern applies across the industry.
The finding also matters because Anthropic’s own safety policy uses accelerated progress as an early-warning trigger. Its Responsible Scaling Policy, version 3.4, identifies a doubling in the pace of progress that is plausibly attributable to automated research or engineering as one trigger for “automated R&D” safeguards. METR said it was plausible that Anthropic’s threshold had been crossed by July 15. That is an assessment of the policy’s trigger, not proof that AI systems can independently build their successors or that a particular risk has materialized.
For readers, the public-evidence gap is part of the story. METR had access to material unavailable to outside researchers, but its report says much of the most informative evidence was redacted. The published estimate can inform discussion, while the limits on disclosure make it harder to assess the underlying assumptions and reproduce the analysis.
Products Worth Considering
Samsung Odyssey G9 49-Inch Curved Dual QHD Gaming Monitor
LG UltraGear 32GX850A-B 32-inch 4K UHD OLED Gaming Monitor
Samsung Odyssey G7 40-Inch WUHD Curved Gaming Monitor (G75F)
As an affiliate, we earn on qualifying purchases.
How METR Measured Progress
METR’s work differs from an evaluation of a finished, publicly available model. The organization assessed internal development data at Anthropic, including results from models that had not been released. Its main measure, AECI, combines scores across benchmarks into a single capability index. The reported acceleration compares the rate of change in that index across two periods; it does not mean every research task, team or stage of model development became 1.9 times faster.
Anthropic’s policy defines a 2x acceleration as making in one year as much progress as would previously have taken two. The policy describes this as a potential early warning, rather than evidence on its own that fully autonomous AI research is underway. Anthropic has said that recursive self-improvement would involve an AI system autonomously designing and developing its successor, and that the company is not there yet.
In a November 2 filing with the U.S. Securities and Exchange Commission, Anthropic said it “may imminently meet” its definition of dramatically accelerated progress. METR’s report offers a separate outside estimate based on company data. Its 1.9x result is higher than the 1.7x figure in Anthropic’s Haiku 5.5 system card; METR attributes the difference mainly to its inclusion of newer, unreleased models.
““We would not be surprised” by an acceleration estimate ranging from 1.3x to 3x.”
— METR
What the Redactions Leave Unknown
METR’s central estimate is not a direct measurement of all AI research and development activity. The report’s conclusions depend on benchmark scores, the selected comparison periods and assumptions about how those scores represent capability progress. METR said varying assumptions about task difficulty moved its estimate from 1.1x to 2.9x. It also offered two possible interpretations of the smaller 1.4x result across developers: Anthropic may have lagged the frontier in 2025, making its later acceleration look larger, or its earlier models may have been less tuned to benchmarks, making the measured acceleration look smaller.
The report says Anthropic redacted a majority of the evidence METR found most informative. The public material therefore does not allow readers to independently check all of the data, interview evidence or reasoning behind the estimate. The planned examination of how Anthropic allocates compute between safety work and capability development was also not completed. METR’s estimate that AI accounts for perhaps 80% of the acceleration should be read as an informed judgment, not a precise, publicly auditable breakdown.
It also remains unclear whether the measured rate will persist, whether similar acceleration is occurring at other labs, and how much it reflects changes in benchmark performance rather than broader research productivity. The report does not establish that Anthropic’s systems can autonomously conduct the full process of designing and developing a successor model.
Policy Thresholds and Further Evidence
The next developments to watch are whether Anthropic provides more detail about its redactions or policy assessment, and whether other independent evaluations find similar changes at other AI developers. The public report does not say that Anthropic has formally declared its policy trigger reached. METR’s conclusion that crossing the threshold by July 15 is plausible should not be treated as that declaration.
Further evidence could clarify how much of the measured acceleration comes from AI assistance, how the rate changes over time and whether the result holds under alternative benchmarks and comparisons. Anthropic’s Responsible Scaling Policy links its accelerated-progress trigger to safeguards, so any formal company assessment of that threshold would be relevant. For now, METR’s report offers a significant but qualified estimate: about twice the previous pace at Anthropic, with substantial uncertainty and limited public access to the underlying evidence.
Key Questions
What did METR conclude about Anthropic’s AI progress?
METR estimated that Anthropic’s capability progress accelerated by about 1.9x after an estimated October 2025 change point, compared with its earlier rate. The organization gave a 95% confidence interval of 1.5x to 2.8x and said a wider range of 1.3x to 3x would not surprise it.
Does the report prove AI is building the next AI by itself?
No. METR assessed AI-assisted research and development, in which people remain involved. Anthropic describes fully autonomous design and development of a successor as recursive self-improvement and has said the company is not there yet.
How much of the acceleration did METR attribute to AI?
METR estimated that perhaps 80% of the acceleration was due to AI. This is the evaluators’ judgment; the report does not provide a fully public, independently checkable breakdown of that contribution.
Why can’t readers verify the estimate fully?
METR said Anthropic redacted a majority of the evidence it found most informative. The assessment included internal benchmarks, training data, surveys and staff interviews, some of which are not available for public review.
Did Anthropic confirm it had crossed its safety-policy threshold?
The source material does not establish that Anthropic formally declared the threshold reached. The company said in a November 2 SEC filing that it “may imminently meet” its definition, while METR said crossing the threshold by July 15 was plausible.
Source: ThorstenMeyerAI.com
