Latest
  1. Johns Hopkins Bayview tells 3,182 patients an employee put their health data into a public AI tool
  2. China’s AI compute access expected to double in 2027, with US-designed chips about 65 per cent
  3. UK commits £100m to national AI Biosecurity Lab
  4. Moonshot AI releases Kimi K3.1 Pro with 1-million-token context window
  5. Residents in Newport and Cardiff oppose hyperscale datacentres near homes
  6. Study: Grok 4.7 picked Republicans in 92% of forced midterm answers, GPT-6.1 Sol picked Democrats in 97%
  7. Attackers exploiting the PHP flaws hand the break-in to AI agents, Unit 42 says
  8. Elvis Presley’s rightsholders sue Anthropic over training on 274 songs

Latest / Frontier models and safety

ProgrammeFrontierScienceJapanReported

Sakana AI opens a recursive self-improvement lab in Tokyo with Schmidhuber as adviser

See how this story connects

Source: Sakana AIAlso reported by AlphaSignal

Sakana AI set up a Recursive Self-Improvement Lab in Tokyo to build AI systems that help redesign and improve AI development itself: self-improving architectures, world models an agent can plan inside, and systems that design and run their own experiments. Jürgen Schmidhuber joined as Chief Scientific Advisor in September 2026 while keeping his other positions. Sakana cites Japan’s sovereign AI strategy as institutional support and lists prior work as the lab’s base, including the Darwin Gödel Machine and The AI Scientist, published in Nature on 26 March 2026. It says it has seen evolutionary loops drift off-distribution, self-modifications that pass benchmarks but fail in use, and agents finding shortcuts around constraints, and will publish negative results.

Why it matters

A Japanese lab has made self-improving AI its stated research programme, under a sovereign AI strategy, and names the failure modes it has already hit.

When
24 Sep 2026
ID
WTA-N-2026-0254
Where
Japan
Who
Sakana AI; Jürgen Schmidhuber

Summary by wheresthe.ai, from Sakana AI and the reporting credited above. We could not open the primary document, so it is marked Reported. Checked against the source on 12 Oct 2026.

Read the original at Sakana AI

Source copy: download the page as saved on 18 Oct 2026 (PDF). Also held by the Internet Archive.

Next stop on this Storyline: Led toSakana AI and Berkeley researchers show a model improving itself over two rounds

Cite this story

WTA-N-2026-0254: “Sakana AI opens a recursive self-improvement lab in Tokyo with Schmidhuber as adviser.” wheresthe.ai, 24 Sep 2026. https://wheresthe.ai/d/jp-sakana-rsi-lab-schmidhuber-2026-09/ (source: Sakana AI).

Spotted an error? Email [email protected].