Automation‑first AI‑safety labs

Organize new AI‑safety organizations around heavy use of AI automation and agentic workflows (evaluations, red‑teaming, data curation, reporting) so a small, lean team can scale safety work against rapidly improving capabilities. These labs prioritize building automated tooling and agentic pipelines as the core product, not as an augmentation to large human teams. — If successful, such labs change who can produce credible safety evaluations, accelerate the pace of safety tooling, and shift regulatory and funding questions toward provenance, auditability, and the governance of automated testing pipelines.