🔍 Read the full analysis: AI Helping Advance AI Development? METR’s Anthropic Report on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- 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 1.9x after October 2025, with AI assistance likely responsible for much of the increase. The nonprofit says most of its most informative evidence was redacted, leaving key details difficult for the public to verify.
METR estimates Anthropic’s AI capability progress accelerated about 1.9 times after October 2025, in an independent assessment published November 5 and based on 11 weeks of access to company data. The nonprofit says AI assistance probably accounts for most of the acceleration, but it also reports that Anthropic redacted a majority of the evidence METR found most informative, limiting what outsiders can check.
METR, a nonprofit that evaluates frontier AI systems, worked with internal Anthropic data from August 12 to October 29, 2026. Its assessment drew on benchmark results for unreleased models, training metrics, staff surveys and about 15 interviews. Roughly 2.5 full-time-equivalent METR staff members conducted the work using Anthropic devices. METR says it received no payment; Anthropic provided laptops and free model tokens and retained the right to redact non-public material.
The central estimate compares Anthropic’s rate of progress on a combined capability measure before and after a break point METR places at October 7, 2025. METR estimates the earlier pace at about 15 index points per year and the later pace at about 29, yielding 1.9 times the previous rate. Its 95% confidence interval is 1.5x to 2.8x. METR says it would not be surprised by an estimate ranging from 1.3x to 3x.
METR attributes perhaps about 80% of the acceleration to AI use, while saying it is confident the change did not come from faster growth in training compute. That is the evaluators’ estimate, not a direct measurement of AI’s share of the work. Anthropic’s own figures, as described in the report, say Claude led 26% of the company’s AI research and development work in August 2026, up from less than 1% in February; about 30,000 agents were working at a time on its main internal platform.
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
How AI Assistance Could Speed Research
The report addresses whether AI systems are beginning to help produce more capable successors, a question with consequences for both the pace of AI development and the time available to manage its risks. METR’s estimate suggests AI assistance may already be contributing materially to Anthropic’s research and engineering productivity. It does not show that models are independently designing and building their successors.
That distinction matters because Anthropic’s own Responsible Scaling Policy has a trigger for automated research and development. Under the policy, a doubling in the rate of progress may count if it is plausibly substantially attributable to research or engineering automation. METR says it is plausible that Anthropic crossed the relevant threshold by July 15, 2026. The report frames the result as a measurement bearing on an early-warning threshold, not as proof of autonomous self-improvement or a demonstrated safety failure.
The findings also matter beyond Anthropic. When a major developer uses AI to accelerate internal work, outside observers may need better access to evaluate how quickly capabilities are changing and whether safeguards are keeping pace. Here, the assessment provides an unusual view inside a company, but the redactions mean readers cannot independently inspect much of the evidence supporting its conclusions.
Products Worth Considering
Bose QuietComfort Ultra Earbuds (2nd Gen) – Wireless Noise‑Cancelling Earbuds with Mic, Immersive Audio, USB‑C Charging, Up to 6 Hours Batte
Samsung Galaxy Tab A11+ 11-Inch Tablet, 6GB RAM, 128GB Storage, Expandable, Gray
Kindle Paperwhite 16GB (2024) – 7 Glare-Free Display, Weeks Battery Life – Black
As an affiliate, we earn on qualifying purchases.
METR’s Internal Data Assessment
Most outside evaluations test publicly available models. METR’s report is different because its staff examined company-internal metrics and unreleased-model benchmark scores, alongside interviews and surveys. The approach offers a view of development progress that external benchmark testing alone may miss, while relying on access and information controlled by the company being assessed.
METR’s 1.9x estimate is for Anthropic’s models, not the whole AI industry. When it examined frontier models across developers, the estimated pace rose from about 17 to 24 index points per year, or 1.4x. METR says the difference could mean Anthropic was behind the frontier in 2025, making its company-specific jump look larger, or that earlier Anthropic models were less tuned to the benchmarks, which could make the measured increase smaller.
The estimate also depends on choices about task difficulty. METR says alternative assumptions move the acceleration estimate from 1.1x to 2.9x, while most other checks remain near 2x. The report compares its result with Anthropic’s Haiku 5.5 system card, which gave an estimate of 1.7x; METR says including newer, unreleased models is the main reason its estimate is higher.
““Perhaps ≈80%” of the current acceleration is due to AI.”
— METR
Evidence and Estimates Still Limited
Most of the evidence METR found most informative was redacted by Anthropic, according to the report. The public therefore cannot fully examine the underlying data or assess how the evaluators reached every conclusion. The report also says a planned check of how Anthropic divides compute between safety work and capability development was dropped.
The 1.9x figure is an estimate based on a benchmark index and a selected break point, not a direct count of all research output. METR reports sensitivity to assumptions about task difficulty, and its comparison across all developers produces a smaller acceleration. The available material does not settle how much of the measured change reflects AI assistance, benchmark fit, Anthropic’s position relative to competitors or other factors.
Anthropic’s agent-use figures are company measurements reported in the source material; they do not establish that agents completed the work autonomously or that they caused the capability trend. The report’s assessment of a possible policy-threshold crossing is also a judgment about plausibility, not confirmation that Anthropic formally declared the threshold met.
Policy Threshold and Further Disclosure
The next points to watch are whether Anthropic determines that it has met its Responsible Scaling Policy trigger and what safeguards or reporting that would require. An SEC filing dated November 2 said the company “may imminently meet” its definition of dramatically accelerated progress; that wording indicates a possibility, not a confirmed threshold crossing.
Further public evidence could clarify the contribution of AI tools, the benchmark trend and how the company allocates resources between capability and safety work. METR’s report does not establish when or whether Anthropic will release additional data, and the extent of future disclosure remains unclear.
Key Questions
What did METR estimate about Anthropic’s AI progress?
METR estimated that Anthropic’s progress on its combined capability measure accelerated 1.9x after a break point it dates to October 7, 2025. Its reported 95% confidence interval is 1.5x to 2.8x.
Does the report say AI is building AI by itself?
No. METR’s assessment concerns AI assistance in human-led research and development. Anthropic defines recursive self-improvement as an AI system autonomously designing and developing its successor, and says that has not happened.
How much of the acceleration did METR attribute to AI?
METR guessed that perhaps about 80% came from AI use. The report presents this as an estimate, not a directly measured proportion.
Why is the estimate difficult to verify?
Anthropic redacted a majority of the evidence METR found most informative. METR also says its result changes under different assumptions about task difficulty, and its all-developer comparison gives a lower estimate of 1.4x.
Did Anthropic confirm that it crossed its policy threshold?
No confirmed crossing is reported. METR says it is plausible the threshold was crossed by July 15, while Anthropic’s November 2 SEC filing said the company “may imminently meet” its definition.
Source: ThorstenMeyerAI.com
