report · 2026-11-05
Independent Assessment of AI R&D Acceleration at Anthropic in October 2026
On this page
Why it matters here
METR reports an August–October 2026 assessment of AI R&D acceleration at Anthropic using internal capability histories, research-input data and staff interviews. It judges that progress on well-specified tasks has accelerated and that AI assistance likely contributes substantially, while distinguishing observed benchmark trends from uncertain causal attribution. Highly autonomous contributions are not yet described as a major source of overall progress. The public account exposes important measurement, access and redaction limits rather than independently demonstrating a recursive improvement loop.
AI-R&D capabilityEvaluation integrity
What to keep in mind
- These are METR's source-level findings and subjective judgments. Important evidence and the full report are non-public or redacted, so this tracker cannot independently reproduce the fit or establish AI's causal contribution.
- AECI combines internal and external benchmarks; snapshot inclusion, internal availability dates, link functions and trend assumptions affect estimates. Its uncertainty does not include every choice of model population, and scores are not a cross-lab research-productivity measure.
- Self-reported employee uplift and informal Slack polls are loose evidence, not matched human-comparator trials or end-to-end acceleration rates. The no-AI counterfactual is uncertain; non-AI causes remain plausible.
- Access was mediated through Anthropic staff, with no query access to unreleased models or employee-like access. The compute-allocation verification component was canceled before substantive conclusions.
- METR took no payment but received devices and token credits; both provider and evaluator analyses used Claude extensively. Disclosures include a declined Anthropic job offer, competitor equity, checks by an unconflicted reviewer and informal pressures concerning future access.
- Only the main AECI chart, attribution schematic and self-reported-uplift chart were visually checked here. The attribution schematic explicitly uses fake data; remaining appendix figures and underlying private data were not audited. No numeric observation, public figure or tracker assessment is added.