Anthropic publishes an index showing Claude leading 26% of its own AI R&D
On 17 September 2026, Anthropic released what it calls an R&D Automation Index, reporting that as of August Claude "leads" 26% of the company's AI research and development work, up from effectively zero in February 2026. The company defined "leads" as the model completing most of a given task end-to-end from a high-level prompt while remaining under human supervision, and said roughly 90% of its R&D is done in collaboration with Claude. Anthropic built the index by cataloguing every type of AI R&D task at the company, rating each against an automation scale developed by Epoch AI and weighting by the staff time each task consumes, and disclosed that about 30,000 agents were doing research and engineering work at any moment, with their actions passing through an online monitor before execution. The company proposed two further measures, covering agent oversight and compute allocation, and stopped short of claiming recursive self-improvement.
Why It Mattered
The question of whether AI systems can meaningfully accelerate the development of their successors had been argued for years with anecdotes and forecasts. This is the first attempt by a frontier lab to answer it with a defined metric, a stated methodology, a time series and a starting point — zero in February, 26% in August. Whatever the number's accuracy, its publication changes the terms of the debate: subsequent claims about automated AI research can be compared against a baseline rather than asserted, and rival labs now face an implicit question about their own figures. The disclosure's limits are part of its historical interest. It is self-reported and unaudited, arriving in the same month Anthropic committed to letting outside evaluators inside the company and publish without approval, which makes the index an early test of whether such measures can be independently verified at all. The definition does the heavy lifting — "leads" means end-to-end task completion under supervision, not autonomy — and the company's explicit refusal to characterise this as recursive self-improvement is itself a considered position that will be quoted either way. The operational detail is equally notable for the record: tens of thousands of concurrent agents inside a frontier lab, routed through a monitoring layer, describes an organisational form that did not exist in 2024 and became the working environment of frontier research within two years.
Who Built It
Anthropic
Applications
- AI Research
- Software Engineering
- Capability Measurement