Ban AI That Builds AI

Propose an explicit legal and regulatory prohibition (or strict licensing regime) on autonomous systems that design, train, or materially improve other AI models without human‑in‑the‑loop oversight. The rule would cover agentic pipelines that perform automated architecture search, automated training runs, or chained agentic development that can self‑iterate without transparent human authorization. — If adopted, such a prohibition would narrow the locus of contestation from abstract 'AGI risk' to a specific, enforceable choke point—autonomous self‑improvement—altering corporate strategy, procurement, and export controls.