AIAnthropic reports Claude now leads ~26% of internal R&D through recursive self-improvement
Anthropic disclosed that Claude handles approximately 26% of the company's model R&D work (up from none earlier in 2026), managing large chunks end-to-end under human supervision with ~90% of R&D involving collaboration.
TLDR
This provides concrete metrics on AI accelerating its own development cycle, raising questions about timelines to fuller recursive self-improvement where models could design successors with minimal human input. It intersects with safety debates and Anthropic's dual emphasis on capabilities and caution, while the company considers a new frontier model release ahead of a potential IPO.
Combined views
—
2 Sources, first seen 1h ago
Anthropic reports Claude now leads ~26% of internal R&D through recursive self-improvement
Anthropic disclosed that Claude handles approximately 26% of the company's model R&D work (up from none earlier in 2026), managing large chunks end-to-end under human supervision with ~90% of R&D involving collaboration.
TLDR
This provides concrete metrics on AI accelerating its own development cycle, raising questions about timelines to fuller recursive self-improvement where models could design successors with minimal human input. It intersects with safety debates and Anthropic's dual emphasis on capabilities and caution, while the company considers a new frontier model release ahead of a potential IPO.