mlsanityData to 14 Oct 2026

UnlearnShield: Shielding Forgotten Privacy against Unlearning Inversion

arXiv 2601.20325 ↗First posted 28 Jan 2026cs.CR cs.CVcited by 0

On the map: LLM safety and security → Machine unlearning

Abstract (arXiv)

Machine unlearning is an emerging technique that aims to remove the influence of specific data from trained models, thereby enhancing privacy protection. However, recent research has uncovered critical privacy vulnerabilities, showing that adversaries can exploit unlearning inversion to reconstruct data that was intended to be erased. Despite the severity of this threat, dedicated defenses remain lacking. To address this gap, we propose UnlearnShield, the first defense specifically tailored to counter unlearning inversion. UnlearnShield introduces directional perturbations in the cosine representation space and regulates them through a constraint module to jointly preserve model accuracy and forgetting efficacy, thereby reducing inversion risk while maintaining utility. Experiments demonstrate that it achieves a good trade-off among privacy protection, accuracy, and forgetting.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Machine unlearning, one of 12 directions in LLM safety and security. Builds on 14 of its 19 references that are in the map. First posted on arXiv 28 Jan 2026.

Builds on

14 of 19 references.

Closest ancestors · 4

Same direction, newest first

  1. 2025Dual-View Inference Attack: Machine Unlearning Amplifies Privacy Exposure18 Dec
  2. 2024Learn What You Want to Unlearn: Unlearning Inversion Attacks against Machine Unlearning4 Apr
  3. 2023SalUn: Empowering Machine Unlearning via Gradient-based Weight Saliency in Both Image Classification and Generation19 Oct
  4. 2023Model Sparsity Can Simplify Machine Unlearning11 Apr

Other references · 10

  1. 2015Deep Residual Learning for Image Recognitioncited by 52,096
  2. 2017Understanding Black-box Predictions via Influence Functionscited by 1,741
  3. 2020Inverting Gradients -- How easy is it to break privacy in federated learning?cited by 712
  4. 2019Eternal Sunshine of the Spotless Net: Selective Forgetting in Deep Networkscited by 618
  5. 2020Amnesiac Machine Learningcited by 319
  6. 2021Evaluating Gradient Inversion Attacks and Defenses in Federated Learningcited by 196
  7. 2020Forgetting Outside the Box: Scrubbing Deep Networks of Information Accessible from Input-Output Observationscited by 180
  8. 2020Provable Defense against Privacy Leakage in Federated Learning from Representation Perspectivecited by 94
  9. 2022Concealing Sensitive Samples against Gradient Leakage in Federated Learningcited by 11
  10. 2024Revisiting Gradient Pruning: A Dual Realization for Defending against Gradient Attackscited by 7

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Revisiting Privacy Leakage in Machine Unlearning: Membership Inference Beyond the Forgotten Set
  2. 2025Recalling The Forgotten Class Memberships: Unlearned Models Can Be Noisy Labelers to Leak Privacy
  3. 2025Towards Benchmarking Privacy Vulnerabilities in Selective Forgetting with Large Language Models
  4. 2025CRFU: Compressive Representation Forgetting Against Privacy Leakage on Machine Unlearning
  5. 2025Rectifying Privacy and Efficacy Measurements in Machine Unlearning: A New Inference Attack Perspective
  6. 2025Reminiscence Attack on Residuals: Exploiting Approximate Machine Unlearning for Privacy
  7. 2026Mitigating Privacy Risk via Forget Set-Free Unlearning
  8. 2024Reconstruction Attacks on Machine Unlearning: Simple Models are Vulnerable