🔍 Read the full analysis: AI Helping Build AI? METR’s Findings And The Redacted Details 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 Anthropic’s progress on well-specified AI tasks accelerated about 1.9x after October 2025, and says AI assistance likely accounts for much of the increase. The assessment draws on internal company data, but Anthropic redacted a majority of the evidence METR considered most informative, limiting public scrutiny.
METR, a nonprofit evaluator of frontier AI systems, estimated on November 5 that Anthropic’s progress on well-specified AI tasks has accelerated to about 1.9 times its earlier pace, with AI tools likely responsible for much of the increase. The assessment is based on internal company information, but METR said Anthropic redacted a majority of the evidence it found most informative, leaving important parts of its analysis unavailable for public review.
METR compared Anthropic’s model capability progress before and after an estimated break point of October 7, 2025. Its measure, AECI, combines results across multiple benchmarks into a single capability index. METR calculated that the index rose by about 15 points per year before the break and about 29 points per year afterward, producing an estimated acceleration of 1.9 times. In METR’s definition, a twofold acceleration means making in one year as much progress as the earlier pace would have produced in two years.
The estimate has uncertainty: METR gave a 95% confidence interval of 1.5 to 2.8 times and said it would not be surprised by a result between 1.3 and 3 times. Alternative assumptions about task difficulty shifted the estimate from 1.1 to 2.9 times, while most of its other checks remained near twofold. When METR included frontier models from developers beyond Anthropic, the estimated acceleration was smaller, at 1.4 times, reflecting a change from 17 to 24 index points per year.
METR said it believes perhaps about 80% of the acceleration came from AI use, while expressing confidence that faster growth in training compute did not explain the increase. That share is an estimate, not a directly measured division of causes. Anthropic’s own figures, as reported in the assessment, say Claude led 26% of its AI research and development work in August 2026, compared with less than 1% in February. Anthropic also reported about 30,000 agents operating at a time on its main internal platform. Those figures describe the company’s own measurement and do not, by themselves, establish how much the agents contributed to model improvements.
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 Faster AI Research Matters
If AI tools help researchers build more capable AI systems, development timelines could shorten and safety work may face greater pressure to keep pace. METR’s finding is relevant because it attempts to measure a change in the rate of progress using company data rather than relying only on public model releases. It does not establish that AI systems are independently designing their own successors: the source describes current work as AI assisting human researchers, with people still setting goals.
The estimate also bears on Anthropic’s own safety rules. Its Responsible Scaling Policy identifies a doubling in progress plausibly attributable to automated research or engineering as one possible trigger for safeguards. METR said it was plausible that the threshold had been crossed by July 15, 2026. Separately, Anthropic wrote in a November 2 securities filing that it “may imminently meet” its definition of dramatically accelerated progress. These statements indicate a potential policy milestone, not a confirmed conclusion that a trigger has been met.
For the public, the redactions create a second issue alongside the result itself: how much confidence can outsiders place in an assessment they cannot fully inspect? METR’s methods and headline estimates are available, but withholding much of the evidence judged most informative makes independent checking harder.
Popular right now
De'Longhi La Specialista Arte Evo Espresso Machine with Grinder
Shark AV2501S AI Ultra Robot Vacuum with Self-Empty Base
Shark AV2501AE AI Robot Vacuum with XL HEPA Self-Empty Base
As an affiliate, we earn on qualifying purchases.
How METR Examined Anthropic
METR’s assessment differs from an evaluation of a finished, publicly released model. Its staff worked with internal Anthropic data, including benchmark scores for unreleased models, training metrics, staff surveys and roughly 15 interviews with key employees. METR said about 2.5 full-time-equivalent staff worked on the assessment from August 12 to October 29, 2026, using Anthropic devices. According to METR, it received no payment; Anthropic supplied laptops and free tokens and retained the right to redact non-public information.
The company’s Responsible Scaling Policy, version 3.4, took effect on July 8, 2026. It sets two possible automated-research triggers: an AI system able to replace Anthropic’s research staff at competitive cost, or a doubling of progress plausibly attributable in substantial part to research or engineering automation. The policy describes the doubling threshold as a potential early warning, not proof that severe risks have arrived. Anthropic has also said it is not yet at the stage of an AI system autonomously developing its own successor and that such an outcome is not inevitable.
METR’s 1.9-times figure is higher than the 1.7-times figure in Anthropic’s Haiku 5.5 system card. METR said the difference largely reflects its inclusion of newer, unreleased models. It also cautioned that the Anthropic-only result could be affected by the company’s position relative to the field in 2025 or by differences in how earlier models performed on benchmarks.
““We would not be surprised” by an acceleration between 1.3x and 3x.”
— METR
What the Redactions Leave Unknown
METR said Anthropic redacted a majority of the evidence it found most informative. The public account therefore does not let readers examine all of the underlying data or fully assess how the company’s redactions affect the conclusions. METR also dropped a planned check of how Anthropic divides compute between safety work and capability development. The assessment does not provide that breakdown.
The 1.9-times result is an estimate built from benchmark trends, and METR itself reports meaningful variation under different assumptions. The figures for Claude’s share of work and the number of active agents come from Anthropic’s own measurement, while the claim that AI explains perhaps 80% of the acceleration is METR’s judgment. The precise causal contribution of AI tools remains uncertain.
It is also unclear from the supplied findings whether Anthropic’s policy threshold has formally been triggered. METR said a crossing by July 15 was plausible; Anthropic’s filing said it may imminently meet its definition. Neither statement, as described here, confirms that the company has formally declared the trigger met or specifies what action would follow.
Whether Anthropic Confirms a Trigger
The next relevant development is whether Anthropic determines that either automated-research trigger in its Responsible Scaling Policy has been met and, if so, what safeguards it applies. Its November 2 filing signals that the company sees a possible near-term threshold, while METR’s assessment provides an outside estimate of the pace of progress. No formal trigger determination is established in the material available here.
Further scrutiny will depend on what additional evidence Anthropic makes public and whether METR or other evaluators can assess the data behind the reported acceleration. Until then, the central finding remains a qualified estimate: progress appears to have accelerated, AI likely contributed substantially, and the evidence needed to independently test that conclusion is partly withheld.
Key Questions
What did METR conclude about Anthropic’s AI progress?
METR estimated that Anthropic’s progress on well-specified tasks accelerated about 1.9 times after an October 2025 break point. It reported a 95% confidence interval of 1.5 to 2.8 times and said results could vary under different assumptions.
Did METR prove AI is building AI autonomously?
No. The assessment describes AI tools helping human researchers with AI development. It does not say that an AI system is independently designing and developing its own successor.
How much of the acceleration did METR attribute to AI?
METR estimated that perhaps about 80% came from AI use. That is its judgment about the likely cause, not a directly measured causal breakdown.
Why can’t the public fully check the findings?
METR said Anthropic redacted a majority of the evidence it considered most informative. The redactions limit public access to the information needed to examine the assessment in detail.
Has Anthropic confirmed that its safety-policy threshold was crossed?
The source material does not establish a formal confirmation. METR said a crossing by July 15 was plausible, and Anthropic said in a November 2 filing that it may imminently meet its definition of dramatically accelerated progress.
Source: ThorstenMeyerAI.com
