Towards Reliable Forgetting: A Survey on Machine Unlearning Verification

With growing demands for privacy, security, and legal compliance (e.g., GDPR), machine unlearning has become a critical technique for ensuring the controllability of learning systems. A central challenge in this area is verifying whether unlearning has been successfully performed. Although unlearning methods are widely studied, verification remains underexplored and lacks a unified framework. This survey addresses the gap by organizing existing methods into behavioral and parametric categories based on the evidence used. It compares representative approaches in terms of assumptions, strengths, and vulnerabilities, and concludes with open problems to support the development of more reliable verification mechanisms.