Anthropic says the model now leads a quarter of its own AI research
On 17th Sep'26 Anthropic published the first results from what it calls an R&D Automation Index, its own attempt to measure how much of its AI research and development the model now leads. The figure is 26% as of August, up from 1% in February, scored on a scale built by Epoch AI that runs from no AI involvement at all to full autonomy. "Leads" sits at level four, where the model completes most of a task from a high-level prompt while a person supervises. It was built by having a Claude agent review a fifth of the research staff each week of July using Slack and internal documentation, producing roughly 15,000 tasks that were sorted into 542 categories, with a second Claude acting as judge.
What it actually means
Where the model and the employee agreed on a task's automation level 59% of the time, while two employees rating the same work agreed only 35% of the time. Your team's own sense of how much the model is doing is less reliable than you think.
The copyable part is the method rather than the number. Take one week, list what your team actually did at task level instead of project level, and score each task by how much of it a model can carry end to end today. Freeze that list, then score it again after the next model release. You will find out where the automation actually landed rather than where the vendor's slide says it should have.
Two of its own people agreed about their work less often than the model agreed with either of them.
59% vs 35%model and employee agreement on the same work, against two employees agreeing with each other
Source: Implicator.ai, reporting the Anthropic Institute index · Sunday 20 September 2026