Anthropic says Claude leads 26% of model R&D work

Anthropic disclosed that its Claude AI model now leads 26% of internal model research and development, a sharp increase from 1% in March 2026. The company stated that Claude participates in over 90% of research workflows, with approximately 30,000 AI agents operating simultaneously on its internal platform. Anthropic characterized this integration as the beginning of a feedback loop rather than full autonomy, noting that human researchers retain final approval authority. The disclosure highlights the rapid adoption of AI-assisted development within frontier labs, where models increasingly contribute to training successor systems. Industry observers note this trend compresses development timelines while complicating traditional safety evaluation methods. Anthropic has not released specific metrics on how AI contributions are validated against human benchmarks. The announcement follows broader industry movement toward automated research pipelines, intensifying debate over whether current governance frameworks can adequately monitor self-referential AI development cycles.