🔍 Read the full analysis: METR Examines AI’s Potential To Speed Up Its Own Development on ThorstenMeyerAI.com
Get smart everyday buys 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 Anthropic improved its AI capabilities on well-specified tasks about 1.9 times faster after October 2025 than before, with AI use likely accounting for much of the acceleration. The nonprofit says most of its most informative evidence was redacted, leaving the public unable to independently assess key parts of the analysis.
METR, a nonprofit that evaluates frontier AI systems, estimates that Anthropic’s AI capabilities advanced about 1.9 times faster after October 2025 than during the preceding period, with AI use likely responsible for most of the increase. The report, published November 5 after 11 weeks examining internal company data, also says Anthropic redacted a majority of the evidence METR considered most informative, limiting what outsiders can verify about a finding tied to concerns that AI could speed its own development.
METR measured progress using an index that combines results across capability benchmarks. It estimates that Anthropic’s score rose by about 29 points per year after a break point dated October 7, 2025, compared with about 15 points per year before then. METR describes the result as roughly two years of historical progress occurring in one year. Its 95% confidence interval for the acceleration is 1.5 to 2.8 times, and the organization says it would not be surprised by a figure from 1.3 to 3 times.
The assessment draws on unreleased-model benchmark scores, training metrics, staff surveys and about 15 interviews, according to METR. About 2.5 full-time-equivalent staff worked on the assessment using Anthropic devices from August 12 to October 29. METR says it took no payment; Anthropic provided laptops and free tokens and retained the right to redact non-public information.
METR estimates that about 80% of the acceleration may be due to AI use, while saying it is confident the increase did not come from faster growth in training compute. The figure is an estimate, not a direct measurement of AI’s share. Anthropic’s own data, as described in the source material, say Claude led 26% of its AI research and development work in August 2026, up from less than 1% in February. The company’s main internal platform had about 30,000 agents working at a time.
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 Assistance and Development Speed
The report addresses whether AI is helping researchers build more capable successor systems, a question with implications for the pace of AI progress and the time available to evaluate risks. If AI tools make research and engineering substantially faster, the same teams may be able to develop and test models more quickly. That possibility could affect how companies and policymakers plan evaluations and safeguards.
METR’s estimate does not show that AI systems are independently designing their successors. The source material describes the current process as AI helping humans develop AI, with people still setting goals. Nor does the estimate establish that acceleration will continue at the same rate. Its importance is as evidence of a possible change in the pace of one company’s research, alongside significant limits on public scrutiny.
Anthropic’s policy uses a doubling in the rate of progress that is plausibly attributable to automated research or engineering as one potential trigger for additional safeguards. METR says it is plausible that the company’s early-warning threshold had been crossed by July 15. That is the evaluators’ interpretation, not confirmation that Anthropic formally determined the threshold was met.
Products Worth Considering
Dell UltraSharp 27 Thunderbolt Hub Monitor (U2726DER), QHD
FLEXISPOT 60-Inch Executive Standing Desk with Drawers and CPU Cabinet
Logitech MX Keys Combo for Business Keyboard and Mouse – Graphite
As an affiliate, we earn on qualifying purchases.
Anthropic’s Acceleration Threshold
Anthropic’s Responsible Scaling Policy version 3.4, effective July 8, 2026, identifies two conditions related to automated research and development. One is a model able to replace the company’s research staff at competitive cost; the other is a doubling of progress within a year compared with the previous baseline, where the increase is plausibly substantially attributable to automated research or engineering.
The policy presents the threshold as an early warning, rather than proof that a particular risk has materialized. Anthropic’s November 2 filing with the U.S. Securities and Exchange Commission said the company “may imminently meet” its definition of dramatically accelerated progress, according to the source material. That statement preceded METR’s report and does not itself confirm that the threshold was reached.
METR’s result is higher than the 1.7x estimate in Anthropic’s Haiku 5.5 system card. METR attributes the difference mainly to including newer models that had not been released. When it compares frontier models from multiple developers rather than Anthropic models alone, it estimates a smaller acceleration: from about 17 to 24 index points per year, or 1.4 times.
““roughly two years of historical progress over the last year””
— 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 contents of those redactions, so readers cannot independently examine all of the underlying data or determine how much the omissions affect the assessment.
The estimate also depends on analytical choices. METR says changing assumptions about task difficulty moves the acceleration estimate from 1.1 to 2.9 times, although most of its other checks remain near 2 times. The comparison across all developers yields 1.4 times, which METR says could reflect Anthropic catching up from a weaker position or differences in how earlier models performed on benchmarks.
METR also dropped a planned examination of how Anthropic divides compute between safety work and capability development. The report therefore does not establish whether increased AI assistance has changed that allocation. The estimated 80% contribution from AI is a judgment by METR, not a directly observed share of research output, and the available information does not show whether the pace will persist.
Further Evidence and Policy Decisions
The next significant evidence would include more detail about the redacted material, additional measurements of Anthropic’s progress, and clearer reporting on how AI tools contribute to research tasks over time. Without those details, outside researchers have limited ability to test METR’s interpretation against the underlying evidence.
Anthropic’s policy and public reporting will also be relevant to whether the company decides its acceleration threshold has been met and what safeguards follow. The available source material does not identify a scheduled public update or a confirmed policy action. For now, METR’s report is an independent estimate based on company access, with its central result qualified by methodological uncertainty and restricted public evidence.
Key Questions
What did METR estimate?
METR estimated that Anthropic’s capability progress on its index accelerated about 1.9 times after October 2025 compared with the earlier rate. Its 95% confidence interval was 1.5 to 2.8 times.
Does the report show that AI is building AI on its own?
No. The report concerns AI tools assisting people with AI research and development. The source material says humans still choose goals and that fully autonomous development of a successor system has not been reached.
How much of the acceleration did METR attribute to AI?
METR estimated that perhaps about 80% may be due to AI use. That is an estimate, not a direct measurement, and the report does not establish that the same contribution will continue.
Why can’t the public fully check the findings?
METR says Anthropic redacted a majority of the evidence it considered most informative. The company also retained the right to redact non-public information during the assessment.
Did METR confirm Anthropic crossed its policy threshold?
No. METR said it was plausible the threshold had been crossed by July 15. Anthropic’s November 2 filing said it “may imminently meet” its definition, but the source material does not confirm a formal determination that the threshold was met.
Source: ThorstenMeyerAI.com
