Quantifying Verification Budgets for Probabilistic Model Identity Testing

This research characterizes the statistical separability required for third-party identity verification of stochastic AI models. It establishes how many independent challenges and repeated responses are necessary to achieve a target AUC when verifying remote model ownership.