Labs Plan to Automate Their R&D

Leading AI companies are intentionally building agentic systems to run internal research and engineering tasks so that AIs — not humans — will design, test, and iterate future models. That shift is already underway (companies report large rises in internal inference compute and public roadmaps toward automated researchers) and will reduce the role of human oversight in model development within a short timeframe. — If true, this changes the regulatory problem: governments will need to regulate organizations and development processes (who holds the keys, auditability, and stop buttons), not just model outputs, because oversight capacity will be structurally diminished.