analiza aiqbench
This research establishes an inherent privacy-audit tradeoff in machine unlearning scenarios where model owners are dishonest but auditors seek to verify data removal without compromising the remaining dataset's confidentiality. The study proves that behavioral queries alone cannot effectively audit convex models under mutual distrust, a finding that extends empirically to non-convex architectures and necessitates more robust cryptographic or architectural solutions for future privacy-preserving pipelines.
Behavioral Audit of Machine Unlearning Has a Privacy Cost
arXiv:2606.14518v1 Announce Type: new Abstract: The removal of learned data from Machine Learning models through Machine Unlearning (MU) has been widely studied; however, there has yet to be an agreed-upon scheme for auditing MU. Existing work has shown that a dishonest model owner can falsify evidence to avoid executing MU, while curious auditors (and adversaries) can infer the privacy-sensitive properties of the model and its training data even with limited access. Yet auditing of MU under mutual distrust between the model owner and the auditor remains unexplored. We provide an information-theoretic proof for this scenario: for convex ML models, a generic audit scheme that relies solely on querying the model for \textit{behavioral} signals cannot identify insufficiently unlearned models without revealing membership information of the retained set. Therefore, auditing MU under the assumption of a dishonest model owner and an honest-but-curious auditor faces an inherent privacy-audit tradeoff. Our empirical results on convex models strongly supports this result, while further experiments demonstrate that this privacy-audit tension persists in non-convex models. Our results call for a more careful consideration of the privacy-audit tension under a realistic auditor threat model, and serve as a foundation for more scrutiny of designs of privacy-preserving audit schemes for the MU pipeline. We also release our code implementation at https://github.com/LiouTang/Behavioral-Unlearn-Audit.
Abstract:The removal of learned data from Machine Learning models through Machine Unlearning (MU) has been widely studied; however, there has yet to be an agreed-upon scheme for auditing MU. Existing work has shown that a dishonest model owner can falsify evidence to avoid executing MU, while curious auditors (and adversaries) can infer the privacy-sensitive properties of the model and its training data even with limited access. Yet auditing of MU under mutual distrust between the model owner and the auditor remains unexplored. We provide an information-theoretic proof for this scenario: for convex ML models, a generic audit scheme that relies solely on querying the model for \textit{behavioral} signals cannot identify insufficiently unlearned models without revealing membership information of the retained set. Therefore, auditing MU under the assumption of a dishonest model owner and an honest-but-curious auditor faces an inherent privacy-audit tradeoff. Our empirical results on convex models strongly supports this result, while further experiments demonstrate that this privacy-audit tension persists in non-convex models. Our results call for a more careful consideration of the privacy-audit tension under a realistic auditor threat model, and serve as a foundation for more scrutiny of designs of privacy-preserving audit schemes for the MU pipeline. We also release our code implementation at this https URL.
Machine Learning (cs.LG)
arXiv:2606.14518 [cs.LG]
(or arXiv:2606.14518v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2606.14518 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 303 cuvinte
Subiecte conexe
Generative Model Unlearning: A Survey through Target Events, Unlearning Operators, and Evaluation Protocols
arXiv · 19 aug. 2026
New framework for auditing machine unlearning
Google Research · 10 iun. 2026
Rethinking Demonstration Unlearning in Imitation Learning for Robotics
arXiv · 24 aug. 2026
Characterizing Privacy-Audit Alignment in Behavioral Audit of Machine Unlearning
arXiv · 9 sept. 2026