TL;DR
Get everyday essentials delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
METR’s November 5 assessment estimates that Anthropic’s model capability progress has accelerated about 1.9x since October 2025, compared with its earlier rate. METR says AI use likely explains most of the acceleration, but Anthropic redacted much of the evidence, limiting independent public scrutiny.
METR, a nonprofit that evaluates frontier AI systems, said on November 5 that Anthropic’s model capability progress is advancing at about 1.9 times its earlier rate, based on an assessment using internal company data. METR estimates that AI use likely accounts for most of the acceleration, but says Anthropic redacted a majority of the evidence it considered most informative.
METR’s estimate compares Anthropic’s progress before and after a change point it dates to October 7, 2025. Using Anthropic’s benchmark-based capability measure, METR estimates the rate rose from about 15 points per year to about 29 points per year. It gives a 95% confidence interval of 1.5 to 2.8 times and says it would not be surprised by an acceleration between 1.3 and 3 times.
The assessment was conducted by about 2.5 full-time-equivalent staff over 11 weeks, from August 12 to October 29, 2026. METR says its researchers reviewed benchmark results for unreleased models, training metrics, surveys and about 15 staff interviews. The work took place on Anthropic devices; the company supplied laptops and free model tokens, while retaining the right to redact non-public information. METR says it received no payment.
METR estimates that perhaps 80% of the acceleration is due to AI use and says it is confident faster growth in training compute did not cause the increase. That is an estimate, not a directly measured division of causes. Separately, Anthropic’s internal measure reportedly showed Claude leading 26% of its AI research and development work in August 2026, up from less than 1% in February, with about 30,000 agents running at a time on its main internal platform.
AI progress at Anthropic What METR found
METR estimates Anthropic’s model capability progress has accelerated, with AI tools likely helping human researchers move faster. Redactions limit how much of the evidence the public can check.
How the estimate was built
METR reviewed Anthropic’s internal research and development data over an 11-week assessment.
Researchers examined benchmark results, training metrics, surveys and staff interviews.
METR estimated about 15 capability points per year before the change point and 29 after.
METR says AI use likely explains most of the acceleration; this is an estimate, not a measured causal share.
What the evidence can show
The estimate concerns Anthropic’s measured progress and AI assisting human researchers.
- Progress at Anthropic appears faster than its earlier pace.
- AI tools likely contribute to researchers’ work.
- METR says faster training compute did not cause the increase.
- The report does not show AI systems autonomously building successors.
- Redactions prevent full public scrutiny of key evidence.
- METR’s estimate does not establish the pace across all developers.
Signals and policy threshold
METR’s estimate and Anthropic’s policy language are related signals, but they are different assessments.
What Faster AI Development Could Mean
The report bears on whether AI tools are helping research teams develop more capable systems at a faster pace. If that pattern continues, companies and policymakers may have less time to assess new capabilities and put safeguards in place between development milestones. METR’s results suggest the question is no longer only theoretical: the organization found evidence of a faster pace inside one major AI developer.
But the estimate does not show that AI systems are autonomously building their own successors. The source describes a process in which AI tools assist human researchers, who still set goals and make decisions. METR’s estimate also applies to Anthropic’s measured progress; when it considers frontier models across developers, it calculates a smaller acceleration, from about 17 to 24 capability points per year, or 1.4 times.
The redactions affect how much outsiders can verify about the result. METR says Anthropic withheld a majority of the evidence it found most informative and a planned check of how the company divides compute between safety and capability work was dropped. Readers therefore have a detailed outside assessment, but not the full underlying record.
Popular right now
De'Longhi La Specialista Touch Semi-Automatic Espresso Machine with Grinder
Hisense 40-Inch Class QD4SR Series Hi-QLED FHD Smart Roku TV (2026 Model)
TCL 43-Inch 4K UHD Smart LED Roku TV (43S435, 2021 Model)
As an affiliate, we earn on qualifying purchases.
Anthropic’s Twofold Progress Threshold
Anthropic’s Responsible Scaling Policy, version 3.4, took effect on July 8, 2026. It lists two possible triggers for what it calls automated research and development: a model capable of replacing the company’s entire research staff at competitive cost, or a doubling in the pace of progress that is plausibly substantially attributable to research or engineering automation.
The policy defines a doubling as making, in one year, as much progress as would otherwise take two years at the baseline rate. Anthropic describes this threshold as a potential early warning, not proof that a serious risk has arrived. METR says it is plausible the threshold may have been crossed by July 15, 2026, but that conclusion depends on how the policy’s standard is interpreted.
In a November 2 filing with the U.S. Securities and Exchange Commission, Anthropic wrote that it “may imminently meet” its definition of dramatically accelerated progress. The filing and METR’s assessment are related signals, but they are not the same determination: the company’s statement is a possibility, while METR’s 1.9x figure is its estimate from a specific dataset and method.
““A 2x acceleration means that 2 years of historical progress is made in 1 year.””
— METR
Evidence Behind the Estimate
Most of the evidence METR found most informative was redacted, so the public cannot fully inspect the data or evaluate every part of the analysis. The available source does not specify exactly which evidence was removed or provide enough detail to independently reproduce the 1.9x estimate.
The estimated contribution from AI use is also uncertain: METR says “perhaps” 80%, rather than reporting a directly observed causal share. It is not clear from the available material how much that estimate would change under alternative assumptions, or how the excluded compute allocation analysis might affect interpretation. The 26% figure for Claude leading internal work is Anthropic’s own measurement, and the source does not define in detail what counts as “leads.”
METR’s result concerns Anthropic and a particular capability index. Its cross-developer estimate is lower, and neither figure establishes that all AI development is accelerating at the same rate. The assessment also does not establish that recursive self-improvement has begun.
Policy Triggers and Further Evidence
The next key question is whether Anthropic determines that its Responsible Scaling Policy threshold has been met and what safeguards or disclosures would follow. The company’s November 2 filing described the possibility as imminent; the available source does not report a final threshold determination or a specific action date.
Further public evidence would help readers judge the estimate, including details about the redactions, the methods used to attribute acceleration to AI, and the compute comparison METR planned but did not complete. Until then, the report offers an outside estimate of faster progress while leaving important parts of its evidence base unavailable for public review.
Key Questions
What did METR find at Anthropic?
METR estimates that Anthropic’s model capability progress has accelerated about 1.9 times since a change point it dates to October 7, 2025, compared with the earlier rate.
Does the report show that AI is building AI on its own?
No. The report concerns AI assisting human research and engineering. The source says humans still choose goals and does not establish that systems autonomously design and develop their successors.
How much of the acceleration does METR attribute to AI?
METR estimates that “perhaps” about 80% is due to AI use. That is an attributed estimate, not a directly measured causal share.
Why is the report’s evidence difficult to check?
METR says Anthropic redacted a majority of the evidence it found most informative. The source also says a planned analysis of how Anthropic allocates compute between safety and capability work was dropped.
Has Anthropic said it crossed its acceleration threshold?
Not in the material provided. A November 2 SEC filing said Anthropic “may imminently meet” its definition of dramatically accelerated progress. METR said a crossing by July 15 was plausible, but that is not a confirmed company determination.
Source: Thorsten Meyer AI
