🔍 Read the full analysis: Does AI Now Help Speed Its Own Development? METR’s Findings 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’s November 5 assessment estimates that Anthropic’s capability progress on specified tasks accelerated about 1.9 times after October 2025, compared with its earlier pace. The nonprofit believes AI use likely accounts for most of the acceleration, but says Anthropic redacted a majority of the evidence it found most informative, limiting public verification.
METR, a nonprofit that evaluates frontier AI systems, estimated on November 5 that Anthropic’s model progress on specified tasks accelerated about 1.9 times after October 2025, compared with its earlier pace. METR says AI use likely accounts for most of the increase, a finding relevant to concerns that AI could speed up the development of more capable systems; however, the report says Anthropic redacted a majority of the evidence METR considered most informative.
METR based its assessment on 11 weeks of work with Anthropic’s internal data, including benchmark results for unreleased models, training metrics, staff surveys and about 15 interviews. The review ran from August 12 to October 29, 2026, with about 2.5 full-time-equivalent staff working on Anthropic devices. METR said it took no payment; Anthropic supplied laptops and free model tokens and retained the right to redact non-public material.
The nonprofit’s central estimate compares a pre-October 2025 rate of about 15 AECI points per year with about 29 points per year afterward. AECI combines results from multiple capability benchmarks. METR estimates the change point at October 7, 2025, and describes the resulting rate as roughly two years of the earlier progress occurring in one year. Its estimate is 1.9x, with a 95% confidence interval of 1.5x to 2.8x; METR said it would not be surprised by a result between 1.3x and 3x.
The report estimates that perhaps about 80% of the acceleration is attributable to AI, while saying METR is confident it did not result from faster growth in training compute. That proportion is METR’s judgment, not a directly measured breakdown. The report also notes that results vary under alternative assumptions: changing assumptions about task difficulty shifts the estimate from 1.1x to 2.9x, while most other checks remain near 2x.
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-Driven Progress Matters
The finding bears on whether AI tools are beginning to shorten the research and engineering work needed to create later AI systems. If that effect grows, companies and safety evaluators may have less time to study new capabilities and respond to risks. But METR’s estimate measures a rate of capability progress on specified tasks; it does not show that an AI system is independently designing its successor, or that every part of AI development is moving at twice its former speed.
The result also connects to Anthropic’s own safety commitments. Its Responsible Scaling Policy sets an automated-R&D trigger around a doubling of progress when that acceleration is plausibly substantially attributable to automated research or engineering. METR says it is plausible the threshold was crossed by July 15, 2026. That is an assessment against a company policy threshold, not independent confirmation that a particular safeguard was triggered or that a specific risk has materialized.
Public scrutiny is a central issue. METR had access to internal records, but readers cannot examine much of the evidence behind its conclusions because of redactions. The report therefore offers a detailed external estimate while leaving outsiders with limited ability to reproduce or independently evaluate its most informative inputs.
Products Worth Considering
Apple MacBook Pro MPHF3LL/A 14.2-Inch Laptop | M2 Pro 12-Core Chip 16GB Unified Memory 1TB SSD | 19-Core GPU Liquid Retina XDR Display Space
LG 27UP850K-W 27-inch 4K UHD IPS Monitor
LG PuriCare AeroHit Smart Air Purifier for Large Rooms up to 1000 sq ft, True HEPA Filter with 360° Airflow, PM1.0 Sensor & LED Air Quality
As an affiliate, we earn on qualifying purchases.
How METR Measured the Acceleration
METR’s acceleration measure compares the current pace of capability improvement with an earlier baseline. In its definition, 2x acceleration means two years of historical progress in one year. It is not a claim that researchers feel twice as productive, or that every stage of development takes half as long.
METR’s estimate uses Anthropic’s AECI, a composite measure based on benchmark performance. When the nonprofit instead examined frontier models across all developers, it estimated a smaller change: progress rose from about 17 to 24 points per year, or 1.4x. METR offered two possible explanations for the difference: Anthropic may have been behind the overall frontier in 2025, or its earlier models may have been less tuned to benchmarks. The report does not establish which explanation is correct.
Anthropic’s own data, as described in the source report, indicate Claude led 26% of the company’s AI research and development work in August 2026, up from less than 1% in February. The company also reported about 30,000 agents working at a time on its main internal platform. Those figures describe Anthropic’s internal measures and do not by themselves show how much of the measured capability acceleration those agents caused.
“A 2x acceleration means that 2 years of historical progress is made in 1 year.”
— METR
Redactions Limit Public Verification
The largest limitation is that Anthropic redacted a majority of the evidence METR found most informative. Although METR reviewed internal data and interviewed staff, the public cannot see much of the underlying material needed to check how the evidence supports the estimate. Anthropic also retained the right to remove non-public information from the assessment.
The estimate is not a direct measure of the full pace of AI development. It tracks benchmark-based capability progress for Anthropic models, and the result changes under some assumptions and when the sample includes frontier models from other developers. METR’s estimate that AI explains about 80% of the acceleration is explicitly a judgment. The available material does not provide a public, independently checkable accounting of the separate contributions from AI tools, staffing, research choices or other factors.
The source material also says METR dropped a planned check of how Anthropic splits compute between safety work and capability development. It is not clear from the available information how that allocation would affect the report’s conclusions. Nor does the assessment establish that Anthropic has reached fully autonomous recursive self-improvement; the source describes humans as still choosing goals and directing work.
Policy Thresholds and Further Evidence
The next key development is whether Anthropic’s own monitoring and safety processes treat the reported pace as meeting its automated-R&D threshold. METR said crossing that threshold by July 15 was plausible, while 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. Neither statement, as presented in the source material, confirms that the threshold has formally been declared met.
Further public evidence would help clarify the estimate: in particular, more detail on the redacted findings, the basis for attributing most of the acceleration to AI, and how the company assesses the relevant policy trigger. Until then, readers should treat the 1.9x figure as METR’s best estimate from an internal-data review, with significant limits on public scrutiny—not as proof that AI systems are autonomously developing their successors.
Key Questions
What did METR estimate?
METR estimated that Anthropic’s capability progress on specified tasks accelerated about 1.9x after October 2025, compared with its earlier pace. The report gives a 95% confidence interval of 1.5x to 2.8x.
Does the report show AI is developing AI without human involvement?
No. The assessment concerns AI helping with research and development. The source material says humans still choose the goals; it does not establish fully autonomous development of a successor model.
How much of the acceleration did METR attribute to AI?
METR estimated that perhaps about 80% was due to AI. The report presents this as an informed judgment, not a directly measured share.
Why is the finding difficult to verify independently?
Anthropic redacted a majority of the evidence METR considered most informative. METR had access to internal records, but much of the material behind its analysis is not available for public review.
Did Anthropic confirm it crossed its safety-policy threshold?
The source material does not confirm that. METR said it was plausible the threshold had been crossed by July 15, while an Anthropic SEC filing said the company may imminently meet its definition of dramatically accelerated progress.
Source: ThorstenMeyerAI.com
