Independent AI Assurance Becomes a Funded Infrastructure Layer

METR's reported increase from $13.6 million in 2024 contributions to roughly $71 million recently, alongside Vals AI's reported $40 million round, signals demand for independent model evaluation and safety gatekeeping. DeepRSI-style longer-horizon recursive-improvement tests could make assurance more relevant to frontier deployment, but evaluator independence, benchmark validity, reproducibility, model access, reporting standards, and resistance to capture remain unresolved.