Anthropic revealed that Claude now leads 26% of internal AI R&D, up from under 1% in February, signaling rapid progress in recursive self-improvement. Approximately 30,000 AI agents operate on its platform, making over 1 billion decisions in August. Most R&D reached AL4, where humans provide goals and Claude executes end-to-end work with supervision. While no tasks are fully autonomous AL5, Anthantic maintains strict safety monitoring, blocking 0.002% of decisions online. This transparency sets an industry benchmark as competitors like OpenAI and Google DeepMind advance automated research capabilities, highlighting RSI as a core investment logic for AI capital expenditures.