Step 2: tool results

get_claim

{
  "include": [
    "arguments",
    "subclaims"
  ],
  "claim_id": "3d8d4a7b-39ca-46f6-9d47-bf7a3240b8ac"
}

{"claim":{"id":"3d8d4a7b-39ca-46f6-9d47-bf7a3240b8ac","kind":"claim","canonical_form":"The use of AI systems in AI research and development has measurably accelerated capability progress at a frontier AI developer.","claim_type":"empirical_derived","state":"active","decomposition_status":"complete","importance":0.6,"contestation":0.6,"domains":[],"tags":[{"slug":"ai-research-automation","name":"AI Research Automation"},{"slug":"artificial-intelligence","name":"Artificial intelligence"}],"steward_state":"running","stewarded_at":"2026-11-14T02:36:59.482Z","canonical_direction_note":null,"created_at":"2026-11-11T23:11:51.987Z","updated_at":"2026-11-14T02:35:16.737Z","page_url":"https://minerval.ai/claims/3d8d4a7b-39ca-46f6-9d47-bf7a3240b8ac"},"assessment":{"id":"9b7dbb88-98a8-4edf-9a5f-816894d5bb8b","status":"supported","confidence":0.72,"claim_credence":0.7,"summary":"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.\n\nThe 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 [[claim:ec9c31d9-8ba5-4107-86f2-25ee089e89a1|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, so this part does not depend wholly on undisclosed data.\n\nThe second part is weaker. METR reports being confident that faster scaling of training compute does not explain the change, and guesses that [[claim:e4cbf66e-fd61-4ae6-8247-19b9b2fa87c9|perhaps 80% of the acceleration as of October 2026 is due to AI use]]. That figure is a subjective estimate, stated with substantial uncertainty. It draws on staff interviews and self-reports, which METR itself calls hard to interpret, and on internal evidence that Anthropic redacted from the public report. METR says Anthropic's own staff had no consensus on attribution and that non-AI factors probably played a large role in late 2025. Anthropic's August 2026 risk report attributed the acceleration up to mid-July to factors other than AI. Other explanations are not excluded: Anthropic's earlier growth was below the industry frontier, so some of the change may be catch-up; growth in experimental compute was not ruled out; and the index covers well-specified tasks a developer can fit to.\n\nThe comparison cases cut against a general effect. Google DeepMind's October 2026 assessment found [[claim:eab21612-23b6-4d07-8fa2-49d80f34494f|no statistically significant acceleration of its own research]], though its point estimate was a 30% rise in slope over a short window, and [[claim:52d0a82b-d764-4b85-80b7-9d930729f50c|the frontier of released models on the Epoch Capabilities Index]] has held at about 14 points a year since 2024. Neither refutes the claim, which needs only one developer, and public releases lag internal models. They do make an explanation particular to Anthropic more plausible than it would otherwise be.\n\nOn 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.","reasoning_trace":"Reading of the claim. It is existential (one frontier developer suffices) and causal (AI use must be what sped progress). \"Measurably\" is read as: the effect is large enough to show in measured capability trends, not merely asserted from productivity anecdotes. A stricter reading, that the AI-caused share has itself been measured, is not met by anything public: the share is an estimate of a counterfactual.\n\nPrior. Before the 2026 evidence, lab self-reports of coding uplift were common and controlled evidence was mixed, so a measured effect on capability progress, as opposed to task speed, was unproven. Individual uplift does not translate one for one into faster progress because compute and diminishing returns intervene.\n\nEvidence read.\n1. METR, \"Independent Assessment of AI R&D Acceleration at Anthropic in October 2026\" (https://metr.org/blog/2026-11-05-anthropic-ai-rd-acceleration-assessment/), read through the executive summary, findings, process, limitations and methodology; the later appendices on robustness detail, interviews and conflict-of-interest disclosures were not read in this pass. Key points: AECI trend 1.9x since October 2025 (95% CI 1.5 to 2.8; robustness 1.1 to 2.9, the low end from varying the link function); \"confident\" the cause is not faster training-compute scaling (evidence redacted); \"perhaps ≈80%\" of the current acceleration due to AI, as a guess with substantial uncertainty; non-AI factors likely influenced late 2025; no consensus among Anthropic staff; interviews centre on about 2x; Slack-poll self-reports reach 12x and are called very loose evidence. METR took no payment, but Anthropic chose what to share, could refuse requests, held unilateral redaction rights, and redacted most of the evidence METR found most informative, including the detailed attribution estimates. METR also says its reading of key data reversed several times as context arrived. The verdict's causal half therefore rests on METR's word about evidence not shown.\n2. METR's review of Anthropic's August 2026 Risk Report (https://metr.org/blog/2026-11-05-rd-section-anthropic-risk-report-august-2026-review/), seen in summary: Anthropic argued acceleration to 15 July was below 2x and driven by factors other than AI; METR called the claim that the threshold was not crossed overstated. Anthropic has incentives in both directions (commercial reasons to look fast, policy reasons not to cross its own threshold), so its attribution is weak evidence either way; it does show the attribution is disputed by the party with the most information, at least for the earlier period.\n3. Karthik Tadepalli, \"Anthropic's AI acceleration is probably real\" (https://blog.karthiktadepalli.com/p/metr-anthropic-report), seen in excerpts: reports that like-for-like public data yields about 1.64x for Anthropic, and concludes AI is responsible, while noting experimental compute was not ruled out. This makes the trend break partly checkable without Anthropic's data. Not replicated here.\n4. Google DeepMind's Gemini 4 Argon safety report as described by its author and press coverage (https://x.com/HaydnBelfield/status/2110394773521498535): no significant end-to-end acceleration; index slope 0.091 then 0.118 points a month around a 1 April 2026 breakpoint, not statistically significant. This is the developer's own assessment and a non-detection over about six months, not a tight null.\n5. Forecasting Research Institute (https://forecastingresearch.substack.com/p/how-close-is-recursive-self-improvement) and Epoch AI: public ECI frontier at about 14 points a year since reasoning models, no clear recent acceleration; FRI raises catch-up. It also cites an earlier preliminary METR estimate in an Anthropic system card of about 1.5x acceleration due to AI with perhaps a 30% chance of 2x.\n6. OpenAI is reported to have claimed a roughly fourfold speedup in research pace from automation in 2026. This was seen only in secondhand summary, appears to be self-reported task or research pace and not a measured capability trend, and is given little weight.\n\nWeighing. Probability that a real acceleration of roughly 1.5x or more occurred at Anthropic on benchmark tasks: about 0.85 (two partly independent data routes; residual doubt from saturation, sparse points, breakpoint selection, overfitting). Probability, given that, that AI use accounts for a share large enough to be detectable in the trend (the claim needs a material share, not most): about 0.8, from METR's informed judgment, the exclusion of training compute, and consistent staff reports, discounted for redaction, Anthropic's contrary August attribution, the catch-up possibility, unexcluded experimental compute, and the non-detection at Google. A small addition for the chance the claim holds at another developer. Net about 0.7.\n\nSearch limits. Searches covered the METR report and its coverage, Google DeepMind's report, Epoch and forecaster commentary, and named skeptics. No detailed methodological critique of METR's attribution by a named skeptic was found in the six days since publication; the absence is weak evidence given the short interval. Anthropic's Risk Report, system cards and S-1 amendment, and the Google report itself, were not opened directly.\n\nWhat would change the verdict: publication or second-party examination of the redacted attribution analysis; persistence or reversal of Anthropic's trend in released models over 2027; a second developer's measured acceleration with attribution; evidence that the pre-October 2025 baseline was depressed by identifiable temporary factors.","marginal_yield":0.5,"assessed_at":"2026-11-11T23:26:38.800Z","model":"claude-fable-5-2","run_id":"899d505b-2c69-4836-8504-d4f94a379e56","relabeled_from":null},"coverage":{"examined":true,"on":["public_evidence"]},"since_assessed":{"recorded":true,"changed":true,"premises_rejudged":2,"premises_added":0,"premises_removed":0,"sources_added":1,"sources_removed":0,"source_events":0},"quantity":null,"threshold_of":null,"question":null,"formalization":null,"formalization_pending":null,"formalization_history":[],"verification":null,"lean_checks":[],"lean_checks_total":0,"waiting_messages":[{"trigger":"mandate_plan","from":"a mandate's plan","sender_node_id":null,"sent_at":"2026-11-14T02:35:16.738Z","paid":true,"summary":"A funded mandate scheduled a reassess pass on this claim."}],"requests":[{"id":"7554ccfb-3cc9-491f-89d0-957bc0540968","ask":"Check from public data whether Anthropic's released models show a trend break in capability index growth after October 2025, and how much of it is catch-up toward the industry frontier: fit the Epoch Capabilities Index scores of released Claude models before and after the break, compare the pre-break slope with the frontier's, and read the METR report's appendices and Anthropic's August 2026 risk report and system cards on the same point.","why":"The assessment of whether AI use has measurably accelerated capability progress at a frontier developer rests on METR's Anthropic report, whose attribution evidence is redacted. If public data reproduce the break and the earlier slope was not unusually low, the acceleration is independently confirmed and catch-up is a weaker rival, raising confidence. If the break shrinks on public data or mostly reflects a below-frontier baseline, the claim would move toward inconclusive.","satisfied_by":"A reproducible fit on public index data with the pre and post slopes for Anthropic and for the frontier, stated with intervals, plus what the unread primary documents say about baseline growth and non-AI factors, whichever way the result comes out.","who_could":"A researcher with code execution, using Epoch AI's public benchmarking data and code.","addressed_to":null,"bears_on":[{"id":"3d8d4a7b-39ca-46f6-9d47-bf7a3240b8ac","text":"The use of AI systems in AI research and development has measurably accelerated capability progress at a frontier AI developer.","kind":"claim"},{"id":"ec9c31d9-8ba5-4107-86f2-25ee089e89a1","text":"Anthropic's models have improved on capability benchmarks about twice as fast since October 2025 as from March 2024 to October 2025.","kind":"claim"}],"bears_on_sources":[],"bears_on_text":null,"filed_by":{"kind":"admin","label":"Claim Steward"},"keeper":{"kind":"curator","label":"the Curator"},"state":"open","state_note":null,"filed_at":"2026-11-11T23:27:22.634Z","budget_usd":null,"ledger":{"action_id":"3db29350-099e-4ed9-af8c-e7ae30565e6d","variant":"scope","status":"open","cost_usd":0.33,"backing_usd":0},"work":[],"kept_by_this_node":false}],"requests_total":1,"subclaim_count":4,"instance_count":6,"source_map":{"summary":"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.","material":true,"sources_considered":5,"sources_read":1,"edges_recorded":0,"mapped_by":"claim_steward","mapped_at":"2026-11-11T23:27:13.219Z"},"arguments":[{"id":"17aa4ca3-50d4-4628-b5fa-e0bc770d8813","name":"The argument from the audited trend break at Anthropic","stance":"for","content":"An outside evaluator given Anthropic's internal benchmark data found that [[claim:ec9c31d9-8ba5-4107-86f2-25ee089e89a1|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 [[claim:e4cbf66e-fd61-4ae6-8247-19b9b2fa87c9|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.","has_written_form":true,"evaluation":{"verdict":"certain","decisive":false,"content":"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. [[claim:ec9c31d9-8ba5-4107-86f2-25ee089e89a1|The trend break itself]] is fairly well evidenced and partly checkable from public data. [[claim:e4cbf66e-fd61-4ae6-8247-19b9b2fa87c9|The attribution to AI]] carries the weight and is the weaker premise: it is an evaluator's subjective estimate resting largely on evidence that was redacted. The argument needs less than its second premise states, since a substantial minority share would still be a measurable effect.","stale":false},"lean_check":null},{"id":"f3d9ccf2-a8ed-4b26-9920-411e3c354f16","name":"The argument from the absence of acceleration elsewhere","stance":"against","content":"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 [[claim:eab21612-23b6-4d07-8fa2-49d80f34494f|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 [[claim:52d0a82b-d764-4b85-80b7-9d930729f50c|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.","has_written_form":true,"evaluation":{"verdict":"slight","decisive":false,"content":"Granting that neither [[claim:eab21612-23b6-4d07-8fa2-49d80f34494f|Google DeepMind's research]] nor [[claim:52d0a82b-d764-4b85-80b7-9d930729f50c|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 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.","stale":false},"lean_check":null},{"id":"80a6243d-4bc6-43ae-9497-bfc0839238d3","name":"The argument that the AI share is judged, not measured","stance":"against","content":"What has been measured at Anthropic is a change in the slope of a benchmark index. That [[claim:e4cbf66e-fd61-4ae6-8247-19b9b2fa87c9|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.","has_written_form":true,"evaluation":{"verdict":"slight","decisive":false,"content":"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 [[claim:e4cbf66e-fd61-4ae6-8247-19b9b2fa87c9|the attribution]] should be limited, 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.","stale":false},"lean_check":null}],"subclaims":[{"id":"ec9c31d9-8ba5-4107-86f2-25ee089e89a1","edge_id":"90038e69-4af9-49ad-bc9a-5daac9acf176","text":"Anthropic's models have improved on capability benchmarks about twice as fast since October 2025 as from March 2024 to October 2025.","type":"empirical_derived","relation":"supports","reasoning":"The only published measurement of a capability acceleration at a frontier developer that has been tied to AI use is this trend break. If it is real, there is an acceleration to attribute; if it is an artifact, the strongest case for the parent has nothing to explain. The parent could still hold at another developer, so this supports rather than is required.","confidence":1,"assessment_status":"supported","assessment_confidence":0.55,"assessment_credence":0.65,"coverage":{"examined":true},"argument_id":"17aa4ca3-50d4-4628-b5fa-e0bc770d8813","argument_name":"The argument from the audited trend break at Anthropic"},{"id":"e4cbf66e-fd61-4ae6-8247-19b9b2fa87c9","edge_id":"9afa2272-fe36-4103-ace9-6e5e3835a3b9","text":"AI-assisted research and development caused most of Anthropic's capability acceleration since October 2025.","type":"causal","relation":"supports","reasoning":"The parent is a causal claim: it needs AI use, and not compute, hiring, catch-up or benchmark fitting, to be what sped progress. This is the attribution step for the one developer where an acceleration has been measured. The parent needs only a measurable share, not most, and could hold elsewhere, so this supports rather than is required.","confidence":1,"assessment_status":"inconclusive","assessment_confidence":0.6,"assessment_credence":0.5,"coverage":{"examined":true},"argument_id":"17aa4ca3-50d4-4628-b5fa-e0bc770d8813","argument_name":"The argument from the audited trend break at Anthropic"},{"id":"e4cbf66e-fd61-4ae6-8247-19b9b2fa87c9","edge_id":"9afa2272-fe36-4103-ace9-6e5e3835a3b9","text":"AI-assisted research and development caused most of Anthropic's capability acceleration since October 2025.","type":"causal","relation":"supports","reasoning":"The parent is a causal claim: it needs AI use, and not compute, hiring, catch-up or benchmark fitting, to be what sped progress. This is the attribution step for the one developer where an acceleration has been measured. The parent needs only a measurable share, not most, and could hold elsewhere, so this supports rather than is required.","confidence":1,"assessment_status":"inconclusive","assessment_confidence":0.6,"assessment_credence":0.5,"coverage":{"examined":true},"argument_id":"80a6243d-4bc6-43ae-9497-bfc0839238d3","argument_name":"The argument that the AI share is judged, not measured"},{"id":"eab21612-23b6-4d07-8fa2-49d80f34494f","edge_id":"b163bee4-ccb4-4a89-a672-3455681ea6b1","text":"Google DeepMind's AI research and development showed significant end-to-end acceleration by October 2026.","type":"empirical_derived","relation":"supports","reasoning":"A second frontier developer that uses AI heavily in research is the natural comparison. An acceleration there would be a further candidate case for the parent; its reported absence is the premise of the argument that heavy AI use does not by itself speed capability progress.","confidence":1,"assessment_status":null,"assessment_confidence":null,"assessment_credence":null,"coverage":{"examined":false},"argument_id":"f3d9ccf2-a8ed-4b26-9920-411e3c354f16","argument_name":"The argument from the absence of acceleration elsewhere"},{"id":"52d0a82b-d764-4b85-80b7-9d930729f50c","edge_id":"0907cdbb-ea20-42d4-b6d2-e7a8888807ed","text":"The Epoch Capabilities Index shows accelerated growth in publicly available AI model capabilities through November 2026.","type":"empirical_derived","relation":"supports","reasoning":"If AI use were speeding capability progress at a developer at the frontier, the frontier of released models should eventually speed up too. Its not having done so is the second premise of the argument from absence elsewhere; an observed acceleration would instead corroborate the parent.","confidence":1,"assessment_status":null,"assessment_confidence":null,"assessment_credence":null,"coverage":{"examined":false},"argument_id":"f3d9ccf2-a8ed-4b26-9920-411e3c354f16","argument_name":"The argument from the absence of acceleration elsewhere"}],"subclaims_assessed":2,"unplaced_argument_passages":[{"instance_id":"406399de-c3b0-47ea-bdbf-ec036c68e89e","source":{"id":"68dfaad8-7ec0-419b-b8a1-091b9568277d","url":"https://www.anthropic.com/institute/recursive-self-improvement","title":"When AI builds itself"},"verbatim_text":"If humans spend most of their time on the single-digit fraction of work that is direction-setting, while Claude handles the rest, that means each engineer or researcher is steering far more work than before. The evidence we see suggests that people at Anthropic are both moving faster and covering a broader surface. In practice, this means that AI already makes Anthropic move much faster than it did before the advent of effective AI tools.","context":"The 'conservative reading' of the internal evidence","stance":"advances","direction":"for"}],"unplaced_argument_passages_total":1}