{"id":"arxiv/2602.01150","url":"https://arxiv.org/abs/2602.01150","tags":["cs.LG","cs.AI","cs.CR","cs.CV","math.OC"],"score":0,"title":"SMI: Statistical Membership Inference for Reliable Unlearned Model Auditing","domain":"cs.LG","source":"arxiv","authors":["Jialong Sun","Zeming Wei","Jiaxuan Zou","Jiacheng Gong","Jie Fu","Chengyang Dong","Heng Xu","Jialong Li","Bo Liu"],"reviews":0,"subject":"cs.LG","summary":"Machine unlearning (MU) is essential for enforcing the right to be forgotten in machine learning systems. A key challenge of MU is how to reliably audit whether a model has truly forgotten specified training data. Membership Inference Attacks (MIAs) are widely used for unlearned model auditing, where samples that evade membership detection are regarded as successfully forgotten. We show this assumption is fundamentally flawed: failed membership inference does not imply true forgetting. We prove that unlearned samples occupy fundamentally different positions in the feature space than…","category":"research","comments":0,"disputed":false,"division":"cs","mentions":0,"discussed":0,"published":"2026-02-01","reactions":{}}