Anthropic has revealed metrics on how Claude participates in creating future AI models for the first time. According to data cited by The Decoder, Claude now "leads» 26% of such work, compared to less than 1% in February.
However, the wording requires caution. The source notes that the overall scale of the work is unclear and the assessment was conducted by Claude itself. Furthermore, the word "leads" signifies less than what this metric might imply.
The publication is significant as an attempt by Anthropic to measure AI participation in its own research, but the presented data is insufficient to equate 26% with the share of fully completed work or to draw conclusions about the model's autonomy.
editorial commentary
Why it matters
A likely consequence is increased interest in metrics regarding AI participation in the development of AI systems. The next observable signal will be a publication from Anthropic with a precise definition of the metric and independent verification. Significant uncertainty remains because only a source summary is available, and the assessment is attributed to Claude itself.