Safety researchers blast labs' sudden embrace of recursive self-improvement

Former OpenAI superalignment member So8res and others criticize AI labs for flipping from dismissing recursive self-improvement as sci-fi to openly racing through dangerous-capability checklists.

2026-09-02 ~ 2026-09-02 · 4 related posts

• AI labs treating dangerous capability list like a checklist, philosopher warns — MattLutzPhi · 2026-09-02
• "Agentic AI lets us accelerate recursive self-improvement": lab rhetoric mocked — MattLutzPhi · 2026-09-02
• Ex-OpenAI researcher mocks labs' U-turn on recursive self-improvement — So8res · 2026-09-02
• RSI was on internal roadmaps as early as end of 2023, claims Harris — harris_edouard · 2026-09-02