As large language models are increasingly adopted in federated learning, protecting user privacy while performing parameter-efficient fine-tuning on distributed private data has become an important challenge. Although clients only share gradients instead of directly uploading raw data, the shared gradients may still leak membership information about training samples. ProjRes (S&P, 2026) further increases this risk: with less information and without accessing model outputs, an attacker can effectively distinguish members from non-members solely based on the projection residual between a candidate representation and the subspace induced by server-observable gradients. Existing defenses against membership inference mostly rely on gradient perturbation or regularization, which can not only degrade model utility but also fail to effectively defend against the membership inference attack introduced by ProjRes, which exploits the geometric structure of gradients. To address this issue, we propose FISGuard, a lightweight defense. Its key idea is to construct and fix a low-dimensional representation subspace using independent public data, thereby restricting the space through which private representations are exposed via gradients while preserving the primary information required for downstream tasks. This substantially reduces the projection-residual discrepancy between members and non-members. We evaluate FISGuard against five representative defense methods across three NLP datasets, two LLMs, and two fine-tuning strategies, Adapter and LoRA. The results show that FISGuard reduces the ProjRes attack AUC to near the random-guessing level of 0.5 in most settings, while maintaining downstream task performance close to that of the undefended model and introducing only limited computational overhead, thereby achieving a favorable privacy--utility trade-off.
Photorealistic 3D face avatars are increasingly deployed as reusable digital assets across applications such as telepresence, animation, and personalized media. At the same time, vision-language models (VLMs) can infer sensitive attributes from rendered images with open-ended semantic reasoning without any fine-tuning. This creates a new privacy challenge: once a 3D face avatar is shared, any of its renderings can be analyzed to extract high-level facial attributes. Existing defenses largely operate in 2D image space and do not address identity-preserving semantic manipulation of 3D facial representations. We propose 3D FaceShell, a framework for steering VLM interpretations of faces rendered from 3D models while preserving geometric fidelity and facial identity. 3D FaceShell augments the original 3D representation with a learnable Gaussian shell that produces subtle, spatially distributed perturbations optimized through multi-view embedding alignment. The perturbations are designed to be visually inconspicuous yet sufficient to redirect VLM-based attribute inference in a view-consistent manner. Extensive experiments on reconstructed celebrity face avatars and multiple black-box VLMs demonstrate that 3D FaceShell significantly increases attribute injection and mismatch rates while maintaining high perceptual similarity and identity consistency. Our results show that it is possible to manipulate VLM-level semantic interpretation of 3D faces without compromising their human-recognizable appearance.
Varun Sharma, Kar Wai Fok, Vrizlynn L. L. Thingcs.CR cs.AI
Explainability is central to building trustworthy AI, yet explanation interfaces can inadvertently provide adversaries with an expanded privacy-related attack surfaces. Recent studies show that advanced membership-inference attacks succeed by exploiting confidence-drop trajectories, induced through attribution-guided perturbations, as discriminative features, rather than directly using confidence scores or explanation vectors. Existing defenses against membership inference fail to directly mitigate such explanation-driven attacks. In this work, we investigate whether, during training, a model's own gradients can be leveraged as defense signals against such attacks, thereby aligning explanation profiles between members and non-members. To this end, we propose a Trajectory-Invariant Explanation Regularization (TIER) defense that penalizes erratic fluctuations in confidence drops simulated through gradient-guided perturbations and simultaneously minimizes the distributional shifts via KL-divergence. Unlike conventional adversarial training, which emphasizes label robustness, our approach targets explanation robustness by enforcing self-consistency through KL-divergence and reducing the variance of confidence drops between members and non-members. Extensive experiments confirm that our method effectively mitigates these attacks, delivering privacy protection while maintaining model utility and explanation fidelity.
Large language models (LLMs) are increasingly fine-tuned on domain-specific datasets that may contain sensitive, dataset-level properties. Recent work has shown that such dataset-level information can be effectively extracted through property inference attacks, posing a confidentiality risk. Existing defenses against these attacks primarily operate by modifying the training data distribution and hence require access to the original data and retraining the model, limiting their applicability to settings where data is unavailable or models are already deployed. In this work, we propose alignment-based defenses for mitigating property inference attacks in LLMs. Our approach reshapes the model's output distribution towards a target property ratio via post-training alignment, without modifying the training data. In particular, we adapt two widely used RLHF frameworks--Direct Preference Optimization (DPO) and Group Relative Policy Optimization (GRPO)--as our defenses by constructing preference pairs and defining a specific reward function respectively. Through comprehensive experiments, we show that our alignment based defenses effectively mitigate property inference attacks while maintaining a strong utility confidentiality tradeoff.