Automated AI research closing fast

A cluster of high‑profile statements (Anthropic/Google leaders) and a wave of recent papers on self‑improving agents suggest that automating portions of the AI research pipeline — neural‑architecture search, skill discovery, perpetual self‑evaluation agents — is moving from speculative to operational within months to a few years. If true, this would accelerate capability growth and compress timelines for governance, procurement, and safety oversight. — If AI systems can meaningfully automate research, it changes who controls R&D, shortens upgrade cycles, and raises urgent policy questions about export controls, procurement rules, and safety testing.