🔍 Read the full analysis: How AI May Be Affecting Its Own Development, According To METR on ThorstenMeyerAI.com
Get the latest gadgets 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 progress on measured AI capabilities has accelerated about 1.9 times since October 2025. METR believes AI use likely accounts for much of the change, but Anthropic redacted much of the evidence it considered most informative, leaving key parts of the assessment difficult for outsiders to verify.
METR, a nonprofit that evaluates frontier AI systems, estimated on November 5 that Anthropic’s measured model capability progress accelerated about 1.9 times after October 2025, compared with its earlier rate. METR said AI use likely accounts for most of the increase, but Anthropic redacted a majority of the evidence METR found most informative, limiting what outsiders can independently assess.
METR’s estimate is based on an 11-week assessment involving internal company data, rather than tests of publicly available models alone. About 2.5 full-time-equivalent staff worked from August 12 to October 29, 2026, using Anthropic devices. The nonprofit said it received no payment; Anthropic supplied laptops and free tokens and retained the right to redact nonpublic information. METR reviewed benchmark results for unreleased models, training metrics, staff surveys and about 15 interviews.
The report’s main measure, the Anthropic Epoch Capabilities Index, combines performance on multiple benchmarks. METR estimated that scores rose by about 15 points per year before a breakpoint around October 7, 2025, and about 29 points per year afterward. It describes the change as 1.9 times the previous pace, with a 95% confidence interval of 1.5 to 2.8 times. The estimate is not a claim that all AI research or development work is twice as fast.
METR said its overall judgment was roughly a doubling and that it “would not be surprised” by a result between 1.3 and 3 times. It estimated that perhaps about 80% of the acceleration is attributable to AI use, while saying it was confident the change did not result from faster growth in training compute. Those are METR’s interpretations of the evidence, not independently confirmed measurements of the share of work done by AI.
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-Assisted R&D Matters
If AI tools help researchers build more capable systems, the development cycle could speed up: new models may support work on later models. That could give companies and regulators less time to evaluate changes and prepare safeguards. METR’s findings offer a quantified estimate of that possibility at one company, rather than evidence that AI has become capable of independently designing its own successor.
The distinction matters for policy. Anthropic’s Responsible Scaling Policy identifies a doubling of progress plausibly attributable to automated research or engineering as an early-warning threshold. METR says it is plausible that this threshold was crossed by July 15, 2026. The assessment could inform scrutiny of how labs track AI involvement in research, but the redactions make it harder for outside researchers to judge the estimate and its implications.
The report also does not establish that the same pace applies across the industry. METR’s comparison of frontier models from multiple developers found a smaller acceleration, about 1.4 times, from an estimated 17 to 24 index points per year. The difference suggests that conclusions can depend on which models and comparison period are included.
Products Worth Considering
Apple MacBook Air 13-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, 12MP Center Stage Camera, Touch ID, Wi-Fi 7, Midnight
Kindle Paperwhite Signature Edition 32GB (2024) – Auto-adjusting front light, wireless charging, weeks of battery life – Metallic Black
LG UltraGear 32GS60QC-B 32-inch Curved QHD Gaming Monitor (180Hz, 1ms, 1000R, AMD FreeSync, HDR10)
As an affiliate, we earn on qualifying purchases.
Anthropic’s Early-Warning Threshold
Anthropic’s Responsible Scaling Policy version 3.4 took effect on July 8, 2026. It describes two possible triggers related to automated research and development: a model able to replace the company’s research staff at competitive cost, or progress doubling in a way plausibly substantially attributable to automated research or engineering. The policy defines doubling as achieving in one year the progress that previously took two years.
Anthropic’s policy calls the threshold a potential early warning, not proof that a danger has arrived. Anthropic has also said it is not yet at recursive self-improvement, which it describes as an AI system fully autonomously designing and developing its successor. In the current picture described by the source material, people still set goals and AI assists their work.
METR’s 1.9-times estimate is higher than the 1.7-times figure in Anthropic’s Haiku 5.5 system card. METR said its estimate includes newer unreleased models, which it considers a main reason for the difference. It also cautioned that Anthropic’s earlier position relative to the frontier, or differences in how well models were tuned to benchmarks, could affect the comparison.
“We would not be surprised by anything from 1.3x to 3x.”
— METR
What the Redactions Leave Unknown
Most of the evidence METR found most informative was redacted, according to the report. Anthropic also dropped a planned check of how it divides compute between safety work and capability development. The public account therefore does not allow a full independent review of the underlying records or that allocation.
The estimate itself depends on methodological choices. METR said varying assumptions about task difficulty moved the acceleration estimate from 1.1 to 2.9 times, although it said most other checks remained near two times. It offered two possible explanations for the lower all-developer estimate: Anthropic may have lagged the frontier in 2025, inflating its apparent subsequent acceleration, or its earlier models may have been less tuned to benchmarks, which could push the estimate down.
Other figures in the source material require care. Anthropic’s own measurement reportedly says Claude led 26% of its AI research and development work in August 2026, compared with under 1% in February, with about 30,000 agents operating at a time on its main internal platform. The precise meaning of “leads,” how the share was measured and how much work those agents completed are not established in the account. The figures should not be read as evidence that AI independently performed 26% of all research.
Further Evidence and Policy Triggers
The next public developments are likely to depend on whether Anthropic or METR releases additional evidence, clarifies the redactions or explains how the company measures AI’s role in research. The source material does not identify a scheduled follow-up assessment or a date for further disclosures.
Anthropic’s policy provides a framework for monitoring whether automated R&D reaches its stated thresholds. The company said in a November 2 SEC filing that it “may imminently meet” its definition of dramatically accelerated progress. METR, for its part, judged that crossing the policy’s doubling threshold by July 15 was plausible. Neither statement establishes that Anthropic has formally determined the threshold was met.
For now, the report supplies an outside estimate and a description of the limits on its verification. A clearer public account of the underlying data, the benchmark choices and the extent of AI involvement would help determine whether the measured acceleration reflects a durable change in development pace or factors specific to Anthropic’s models and evaluation methods.
Key Questions
What did METR estimate?
METR estimated that Anthropic’s measured capability progress accelerated by about 1.9 times after a breakpoint around October 7, 2025. Its confidence interval was 1.5 to 2.8 times, and it said a broader range of 1.3 to 3 times would not surprise it.
Does the report show that AI is developing its own successor?
No. The assessment concerns AI assisting people with research and development. Anthropic says it is not yet at a stage where an AI system fully autonomously designs and develops its successor.
How much of the acceleration did METR attribute to AI?
METR estimated that perhaps about 80% of the acceleration came from AI use. This is the nonprofit’s judgment based on the assessment, not a direct public measurement of the share of work completed by AI.
Why is the report difficult to verify independently?
Anthropic redacted a majority of the evidence METR described as most informative. METR also did not complete a planned check of how compute was divided between safety and capability work.
Did METR find the same acceleration across the AI industry?
No. When METR compared frontier models across developers, it estimated a smaller increase of about 1.4 times. The report focuses on Anthropic, and its results do not establish that other companies experienced the same change.
Source: ThorstenMeyerAI.com
