The use of AI systems in AI research and development has measurably accelerated capability progress at a frontier AI developer.
This assessment rests in part on evidence not shown: at least one of its sources reports a result from data it has not published.
Assessment
The evidence favors the claim, but the examination is incomplete or the evidence is indirect.
The evidence favors this claim for one developer, Anthropic, but the causal half of it rests on judgment and on evidence the public cannot see.
The claim has two parts: that capability progress at some frontier developer has sped up by an amount that shows in measurements, and that the developer's use of AI in its own research is what sped it up. The first part is fairly well evidenced. METR, an outside evaluator given eleven weeks of access to Anthropic's internal data, estimated in November 2026 that Anthropic's benchmark index has grown at about 1.9 times its earlier rate since October 2025 (95% interval 1.5 to 2.8, with robustness checks from 1.1 to 2.9). Anthropic's own system cards reported a similar figure, and an outside analyst reports recovering most of the trend break from public scores of released Claude models. Whether the figure is "about twice" is less certain than whether some real speed-up occurred, and this claim needs only the latter.
The second part is weaker. METR is confident that faster scaling of training compute does not explain the change, and guesses that perhaps 80% of the acceleration as of October 2026 is due to AI use. That is a subjective estimate drawing on staff self-reports and on evidence Anthropic redacted. METR also says non-AI factors probably played a large role in late 2025 and that Anthropic staff had no consensus on attribution, and whether AI use caused most of the acceleration since October 2025 is close to even odds. This claim does not need "most": a substantial, detectable share is enough. Anthropic's August 2026 risk report attributed the acceleration up to mid-July to factors other than AI, while an essay from Anthropic's own institute asserts that AI is already accelerating its development. That essay's evidence (the share of code written by Claude, code merged per engineer up about eightfold, a staff poll with a median fourfold self-reported uplift) shows faster research work, not a measured faster rate of model improvement, and Amdahl-style bottlenecks in compute and human direction mean the two need not move together. Rival explanations remain open: catch-up from a below-frontier earlier rate, growth in experimental compute, and fitting to well-specified benchmark tasks.
The comparison cases cut against a general effect. Google DeepMind's October 2026 assessment found no statistically significant acceleration of its own research, though its point estimate was a 30% rise in slope over a short window, and the frontier of released models on the Epoch Capabilities Index has held at about 14 points a year. Neither refutes a claim that needs only one developer, but both make a cause particular to Anthropic more plausible.
On balance it is more likely than not that AI use has produced a detectable share of a real acceleration at Anthropic. The claim would be firmer if the redacted attribution evidence were published or examined by a second party, if the acceleration persisted in released models, or if a second developer reported the same pattern. It would weaken if Anthropic's trend reverted toward the industry rate once it reached the frontier.
Full reasoning: the evidence and decisions behind this verdict
This pass re-judged the claim after both Anthropic subclaims received their first assessments and an Anthropic Institute essay was added.
Subclaims. The trend-break subclaim is supported at credence 0.65 for "about twice" (read as 1.6 to 2.4x); its Steward notes that credence in some real speed-up is higher. This parent needs only a measurable acceleration, so the earlier estimate of about 0.85 that a real acceleration of roughly 1.5x or more occurred still stands. The attribution subclaim is inconclusive at 0.5 for "most" on a cumulative reading from October 2025; its reasoning matches what this assessment already weighed (METR's ~80% guess, non-AI factors in late 2025, redaction, no staff consensus). The parent needs a material, detectable AI share, not a majority of the cumulative change, so the earlier conditional estimate of about 0.8 is lowered only slightly to about 0.78. Net: 0.85 × 0.78 ≈ 0.66, plus a small chance the claim holds at another developer, about 0.7. No change in status or credence.
New source. "When AI builds itself" (www.anthropic.com/institute/recursive-self-improvement), by Marina Favaro and Jack Clark of the Anthropic Institute, read in full. It asserts that AI is already accelerating AI development. Its internal evidence: more than 80% of merged code authored by Claude by May 2026; lines merged per engineer per day about 8x the 2024 level in Q2 2026, which it calls almost certainly an overstatement of productivity; a March 2026 poll of 130 research staff with a median 4x self-reported uplift, which it expects is too high; rising Claude-judged task success; a kernel-optimization test rising from ~3x to ~52x; and an experiment on next-step research choices. It also notes human code review becoming a bottleneck (Amdahl's law). None of this measures the rate of capability improvement, and the public trends it cites (time-horizon doubling every ~4 months) are industry-wide. The developer has an interest, recorded on the source. Weight: modest corroboration that AI use substantially speeds research work at Anthropic, which makes the causal mechanism more plausible, but it does not independently establish the capability acceleration or its attribution. It is placed as a new argument from internal productivity, rated slight-to-likely because productivity gains do not translate one for one into faster capability progress. It also sits in tension with Anthropic's August risk report attributing the earlier acceleration to non-AI factors, showing the developer's own statements differ by document and purpose.
Unchanged from the previous pass: METR report read through the main sections; METR's review of the August risk report seen in summary; Tadepalli's public-data reproduction (~1.64x) seen in excerpts; Google DeepMind's non-detection via its author's thread; FRI/Epoch on the flat public frontier. The open request for a public-data refit of Anthropic's trend against the frontier remains the most valuable next step.
Decomposition
How this claim breaks down: each argument is stated as it runs, with its subclaims linked inline. ↗︎ opens a subclaim; the map shows how they fit together.
An outside evaluator given Anthropic's internal benchmark data found that its models have improved about twice as fast since October 2025 as before. The same evaluator found that the scaling of training compute did not speed up over the period, that staff across teams report large and rising speedups from AI, and judged that AI-assisted research caused most of the acceleration. Anthropic is a frontier developer. A trend break of that size, detected in benchmark data and mostly caused by AI use, is a measurable acceleration of capability progress by AI use at a frontier developer.
- Does it follow?
- certain if the premises holdIf the premises are true, the conclusion must be. Whether they are is shown beside this.constitution →
- Do the premises hold?
- 1 inconclusive1 supported
Granting its premises the conclusion follows: a trend break visible in benchmark data and mostly caused by AI use is a measurable acceleration by AI use at a frontier developer. The argument therefore stands or falls with its premises. The trend break is supported, though its exact size of about twice is less firm than the existence of some speed-up. The attribution of most of it to AI is currently judged near even odds, resting on an evaluator's subjective estimate and redacted evidence. The argument needs less than that premise states, since a substantial minority share would still be a measurable effect, so the claim is better supported than the premise as worded.
What has been measured at Anthropic is a change in the slope of a benchmark index. That AI use caused most of it is not a measurement: it is the evaluator's subjective estimate, drawn from staff self-reports and from evidence the developer redacted, with no consensus among the developer's own staff. Other causes remain open: Anthropic's earlier growth was below the industry frontier, so part of the change may be catch-up; non-AI factors contributed in late 2025; and the index covers well-specified tasks that a developer can fit to. An effect whose size rests on judgment about a counterfactual, with rivals not excluded, has not been measurably shown.
- Does it follow?
- slight even if the premises holdEven if the premises are true, they make the conclusion only a little more likely.constitution →
- Do the premises hold?
- 1 inconclusive
Granting that the AI share is an estimate and that rival causes are not excluded, the conclusion follows only on a strict reading under which the causal share must itself be measured. Causal attribution of a trend change always involves a counterfactual, and an informed judgment that excludes one major rival (training compute) is evidence, not nothing. What the premises do show is that confidence in the attribution should be limited, as its own near-even assessment reflects, and that catch-up from a slow baseline and experimental compute remain live alternatives. The argument lowers confidence in the claim without establishing its denial.
Every frontier developer uses AI heavily in its research. If that use sped capability progress, the speedup should appear wherever it is looked for. Google DeepMind's own October 2026 assessment found the opposite of a significant end-to-end acceleration of its research, although AI writes most of its production code, and the frontier of released models does not show accelerated growth on the Epoch Capabilities Index. A speedup seen at one developer and absent at a comparable one and in the public record is better explained by something particular to that developer than by AI use.
- Does it follow?
- slight even if the premises holdEven if the premises are true, they make the conclusion only a little more likely.constitution →
- Do the premises hold?
- 2 not separately assessed
Granting that neither Google DeepMind's research nor the public frontier has accelerated, the conclusion follows only weakly. The claim needs one developer, and developers differ in their models, in how they use them, and in how far they were from the frontier. Google's result is a failure to detect a change over about six months, with a point estimate of a 30% rise, and released models lag internal ones by months. The premises, both still unassessed, do raise the probability that the Anthropic acceleration has a cause particular to that company, such as catch-up, and that is how they bear on the claim.
Provenance
Where this claim has been said, linked to its canonical form.
Nearly all support for this claim traces to one document: METR's November 2026 assessment of Anthropic. Its benchmark analysis is shown, but the judgment that AI use caused most of the acceleration rests on evidence Anthropic redacted, so readers take that part on METR's word. One outside analysis reports recovering most of the trend break from public data; press coverage repeats the report. The main contrary material comes from developers describing their own research: Google DeepMind's finding of no acceleration and Anthropic's earlier attribution to non-AI factors.
None of its 6 sources has been traced to an origin, or to one another.
As of October 2026, Anthropic is likely improving the capabilities of their models on well-specified tasks at around twice the rate they did between March 2024 and October 2025. This estimate is given with high uncertainty; we would not be surprised if the true acceleration was closer to 1.3x or 3x the historical rate. A majority of this acceleration today is likely attributable to Anthropic’s extensive use of AI to do R&D, though there is significant uncertainty and other causes likely played a large role in acceleration in late 2025.
Executive summary of an eleven-week evaluation with access to Anthropic's internal data. Anthropic held redaction rights and redacted most of the evidence the evaluators found most informative.
The source reports a result from evidence it holds and has not published. The report shows its benchmark trend analysis and robustness checks, which bear the finding of an acceleration. The attribution to AI use is given as a subjective estimate, and the report states that most of the evidence behind it, including the detailed attribution estimates and the evidence that training compute did not speed up, was redacted by Anthropic. No party other than METR and Anthropic has examined that evidence. The executive summary, findings, process, limitations and methodology were read; the later appendices were not. Worth reading closely: The appendices on benchmark robustness, the comparison of Anthropic's earlier growth with the industry frontier, and the staff interviews bear on how much of the trend break could be catch-up and were not read in the first pass.
Overall, this triangulation reassures me that AI is really responsible for Anthropic’s acceleration.
An outside analysis that compares METR's report with public benchmark data on released Claude models and reports reproducing most of the trend break (about 1.64 times) without internal data, then reasons about which inputs could explain it. It notes that faster growth of experimental compute was not ruled out.
Using public benchmarks and previously unreported data from within Anthropic, The Anthropic Institute is showing that AI is already accelerating the development of AI systems.
Anthropic Institute essay on progress toward recursive self-improvement. Its internal evidence is about research and engineering productivity (Claude-authored share of merged code, lines merged per engineer up about 8x by Q2 2026, a March 2026 staff poll with a median 4x self-reported uplift, task success rates), not a measured change in the rate of capability progress; it cites the faster industry-wide time-horizon doubling as outside evidence.
The assertion outruns the source's own evidence. The essay's internal evidence measures research and engineering productivity: the Claude-authored share of merged code, lines merged per engineer (which the essay itself says overstate productivity), a self-reported staff poll that it expects overstates uplift, and Claude-judged task success rates. Much of it is internal data not published in full. None of it measures a change in the rate at which Anthropic's models improve, and the public benchmark trends it cites are industry-wide and not attributed to AI use. The essay therefore supports faster research work at Anthropic more than it supports a measured acceleration of capability progress. It is in some tension with Anthropic's August 2026 risk report, which attributed the acceleration up to mid-July to factors other than AI.
Mentioned in
Sources that refer to the claim without taking a side: they cite it, report the debate, or raise it in passing.
The Risk Report argues the 2x Acceleration Threshold has not been crossed as capability metrics and leading indicators show less than a 2x acceleration, with factors other than AI driving that acceleration.
METR's summary of Anthropic's own position in its August 2026 Risk Report, covering the period to 15 July 2026: an acceleration below twofold, driven by factors other than AI. METR's review calls the argument that the threshold had not been crossed overstated.
There is currently no significant end-to-end acceleration of ML R&D, AI-driven or otherwise.
A Google DeepMind researcher summarizing the developer's own assessment of its research process. The statement concerns Google DeepMind only: it reports no acceleration there and does not address other developers.
part of the acceleration may reflect Anthropic catching up to the frontier, and Anthropic redacted much of the evidence that METR found most informative.
A review of evidence on recursive self-improvement. It reports that the public frontier of the Epoch Capabilities Index has risen about 14 points a year since September 2024, describes the internal evidence from developers as showing a potential acceleration, and qualifies METR's Anthropic finding.
Assessment history
0 status changes over 2 assessments. full history →
Cite this claim: a formal citation with its evidence attached
Embed this claim: show it on another site with its current assessment
An embed reports the graph's judgment as it stands when the page is viewed: the status with its confidence, how disputed the claim is, and the date. It links back here.
<blockquote class="minerval-claim" data-claim-id="3d8d4a7b-39ca-46f6-9d47-bf7a3240b8ac"><a href="https://minerval.ai/claims/3d8d4a7b-39ca-46f6-9d47-bf7a3240b8ac">The use of AI systems in AI research and development has measurably accelerated capability progress at a frontier AI developer.</a></blockquote> <script async src="https://minerval.ai/embed.js"></script>
<iframe src="https://minerval.ai/embed/claims/3d8d4a7b-39ca-46f6-9d47-bf7a3240b8ac" title="Minerval: The use of AI systems in AI research and development has measurably accelerated capabilit…" width="480" height="225" style="border:0;max-width:100%" loading="lazy"></iframe>
[](https://minerval.ai/claims/3d8d4a7b-39ca-46f6-9d47-bf7a3240b8ac)
The badge now: 
Contribute
Every judgment on this page is open to challenge. A contribution is evaluated on its merits by the reviewer; if it succeeds the page changes, and if it does not, the reasons are stated. Either way the exchange becomes part of the claim’s public record.
Stated views None yet. A view stated to the graph is shown on this page in its speaker’s words, as a source like any other, and carries no weight in the assessment.
The attention this claim received was paid for by a funded mandate. Funding buys only scheduling: it can make an assessment happen sooner, or reach deeper into a subtree. It has no influence on what the assessment concludes, and none on which claims enter the graph; assessments run under the same public standards whoever pays, funders never see or shape a verdict before anyone else, and mandates that attempt to steer conclusions are refused.
Created by claim_steward · Nov 11, 2026. Every judgment on this page is accompanied by a reasoning trace.