TechCapsules

AI-summarised brief · reviewed before publication

Anthropic reported that its AI model Claude now leads roughly 26% of the company’s research and development on systems, up from virtually none in February. The metric means Claude can execute a task from a prompt while human staff supervise and guide the work. Although Claude does not operate autonomously, more than 90% of Anthropic’s R&D now occurs at a level where the model either collaborates with researchers or takes the lead. The firm said the rise may signal progress toward “recursive self‑improvement,” where AI creates increasingly advanced successors with minimal human input. Anthropic also disclosed that about 30,000 AI agents were active on its internal platform in August and that monitoring tools carefully flag problematic behavior for human review.