Adaptive Auditing Framework for AI Systems with Anytime-Valid Guarantees

This hypothesis testing framework allows for statistically rigorous auditing of generative AI systems even when sampling and stopping decisions are made mid-collection. It addresses the limitations of small-sample, opportunistic testing where classical statistical assumptions are typically violated.