Researchers have introduced a new auditing framework to address the statistical limitations of verifying machine unlearning. The method detects local data anomalies more effectively than traditional testing to ensure AI compliance and privacy.