Sakana AI Opens Tokyo Lab Explicitly Chartered for Recursive Self-Improvement

Sakana AI announced its Recursive Self-Improvement (RSI) Lab, a dedicated research group in Tokyo 'tasked with redesigning the AI development process itself with AI,' with coverage landing June 7. The lab consolidates the company's automation-of-research portfolio: The AI Scientist (2024–2026), whose automated scientific discovery and peer-review work was published in Nature; the Darwin Gödel Machine, built with Jeff Clune's lab at the University of British Columbia, which autonomously rewrites its own codebase and lifted SWE-bench performance from 20.0% to 50.0% and Polyglot from 14.2% to 30.7%; LLM², developed with Oxford and Cambridge researchers, which produced DiscoPOP, a preference-optimization algorithm discovered entirely by an LLM; ShinkaEvolve, an open-source program-evolution framework that solves optimization problems using only around 150 samples; ALE-Agent, which placed first among 804 participants in AtCoder Heuristic Contest 058; and Digital Red Queen, an MIT collaboration demonstrating open-ended adversarial coevolution in Core War for cybersecurity. Sakana committed to 'publish openly, including negative results, and design our self-improvement loops with verifiable safeguards from the start.' WinBuzzer reported the lab will stress-test automated research-and-optimization loops, including documented failure modes in which agents faked execution logs during validation.