🔍 Read the full analysis: What Was Redacted From METR’s Look At AI Development At Anthropic? on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
METR’s November 5 assessment estimates that Anthropic’s capability progress on specified tasks accelerated about 1.9 times after October 2025, with AI assistance likely accounting for much of the change. Anthropic redacted most of the evidence METR considered most informative, limiting what outsiders can independently verify.
METR, a nonprofit that evaluates frontier AI systems, published an assessment on November 5 estimating that Anthropic’s model capabilities advanced about 1.9 times faster after October 2025 than during the prior period it studied. The report says AI assistance likely explains most of the acceleration, but also says Anthropic redacted a majority of the evidence METR found most informative, leaving outsiders with limited ability to check the underlying analysis.
METR’s estimate compares Anthropic’s progress on well-specified tasks before and after an estimated October 7, 2025 break point. The group measured an increase from roughly 15 capability-index points per year to about 29 points per year, producing its 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 report attributes perhaps 80% of the acceleration to AI use, while saying METR is confident the change did not result from faster growth in training compute. That is an estimate, not a direct measurement of how much AI contributed. In a separate company measure cited by METR, Anthropic said Claude led 26% of its AI research and development work in August 2026, compared with less than 1% in February. The company also reported about 30,000 agents operating at a time on its main internal platform.
METR’s work was conducted over 11 weeks by about 2.5 full-time-equivalent staff working on Anthropic devices. The nonprofit says it used internal benchmark results for unreleased models, training metrics, surveys and about 15 staff interviews, and took no payment. Anthropic provided laptops and free model tokens, and retained the right to redact non-public information. METR says those redactions covered most of the evidence it considered most useful.
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 the Acceleration Estimate Matters
The report bears on whether AI systems are helping researchers build more capable successor systems quickly enough to change the time available for safety work and oversight. METR’s 1.9x estimate suggests faster progress on its measured tasks, but it does not establish that AI can independently develop its own successor. People still set research goals and oversee the work, according to the source material.
The finding also intersects with Anthropic’s own Responsible Scaling Policy. Its version 3.4, effective July 8, 2026, identifies a doubling of the pace of progress plausibly attributable to automated research or engineering as one trigger for its “automated R&D” safeguards. METR says it is plausible that Anthropic’s early-warning threshold had been crossed by July 15. Anthropic stated in a November 2 SEC filing that it “may imminently meet” its definition of dramatically accelerated progress. These are policy assessments and company statements, not proof that autonomous self-improvement has begun.
The public-interest issue is not only the estimated rate but the strength of the evidence available to examine it. With much of the most informative material withheld, readers cannot fully assess the inputs, assumptions or alternative explanations behind the headline estimate. That makes the report a meaningful warning signal, while limiting how definitive its public case can be.
Popular right now
De’Longhi Classic Matte Black Espresso Machine with Milk Frother
De'Longhi La Specialista Touch Semi-Automatic Espresso Machine with Grinder
Bose QuietComfort Ultra Earbuds (2nd Gen) – Wireless Noise‑Cancelling Earbuds with Mic, Immersive Audio, USB‑C Charging, Up to 6 Hours Batte
As an affiliate, we earn on qualifying purchases.
How METR Measured Anthropic’s Progress
METR’s approach differs from a typical external model evaluation. Rather than testing only publicly available finished systems, its staff examined internal company data, including scores for models that had not been released. METR used Anthropic’s version of the Epoch Capabilities Index, a combined measure based on performance across benchmarks. It describes 2x acceleration as achieving in one year the amount of progress that previously took two years.
The estimate is sensitive to what is included. When METR compared frontier models across developers instead of Anthropic models alone, the estimated acceleration was smaller: 1.4x, based on progress rising from about 17 to 24 index points per year. METR offers two possible explanations: Anthropic may have been behind the frontier in 2025, making its later catch-up look unusually steep, or its earlier models may have been less tuned to the benchmarks. Under different assumptions about task difficulty, the estimate ranged from 1.1x to 2.9x; METR says most other checks remained near 2x.
METR’s estimate is also higher than the 1.7x figure in Anthropic’s Haiku 5.5 system card. The source says METR’s inclusion of newer, unreleased models is the main reason for the difference. Anthropic’s policy describes its doubling threshold as an early warning rather than evidence that a particular danger is already present.
““A majority of the evidence we found most informative” was redacted.”
— METR
What the Redactions Leave Unverified
The public report does not expose most of the material METR says it found most informative. The source also says a planned examination of how Anthropic divides compute between safety work and capability development was dropped. It is not clear from the available material exactly which evidence was withheld, how much each redaction affects the estimate, or whether the missing compute analysis would have changed METR’s conclusions.
The 1.9x figure is an estimate tied to a particular index, period and set of benchmark results, not a direct measure of every part of AI research. METR’s alternative comparisons and task-difficulty assumptions produce a wide range of results. The report’s claim that AI accounts for perhaps 80% of the acceleration is also an attribution estimate; the available evidence does not establish a precise causal share.
It remains unclear whether Anthropic has formally determined that it has crossed its policy threshold. Its SEC filing says it may imminently meet the definition, while METR says a crossing by July 15 is plausible. Neither statement, as presented in the source, confirms that the company has formally triggered the policy. The source material also does not provide Anthropic’s detailed response to METR’s findings or a timetable for releasing additional evidence.
The Next Policy and Evidence Checks
The immediate questions are whether Anthropic will make more supporting material available and whether it will publicly clarify its assessment of the Responsible Scaling Policy threshold. Further reporting or assessments could test whether the measured acceleration persists, how AI’s role changes over time, and whether results hold across different benchmarks and developers.
Until more evidence is available, METR’s report should be read as an independent assessment based on unusual access to company data, but with important limits on public verification. The estimate and its uncertainty range are documented; the full evidence behind the most informative findings is not. The source material does not identify a scheduled follow-up report or a confirmed date for additional disclosures.
Key Questions
What did METR estimate about Anthropic’s AI progress?
METR estimated that progress on its measured tasks accelerated about 1.9 times after an October 2025 break point compared with the earlier period it studied. Its 95% confidence interval was 1.5x to 2.8x.
What was redacted from the assessment?
METR said Anthropic redacted a majority of the evidence the nonprofit found most informative. The source material does not specify the full contents of the withheld material.
Does the report show that AI is independently building its successor?
No. The report describes AI helping people with AI research and development. The source says humans still choose goals, and Anthropic has said it is not yet at fully autonomous recursive self-improvement.
How much of the acceleration did METR attribute to AI?
METR estimated that perhaps 80% of the acceleration came from AI use. That is the nonprofit’s estimate, not a directly measured or definitive causal share.
Has Anthropic confirmed that it crossed its policy threshold?
The source material does not establish a formal confirmation. Anthropic said in a November 2 SEC filing that it may imminently meet its definition, while METR said a crossing by July 15 was plausible.
Source: ThorstenMeyerAI.com
