Personalized Privacy Control in LLMs via Attention Head Intervention

• URL: http://arxiv.org/abs/2608.21209v1
• Date: Fri, 21 Aug 2026 15:22:20 GMT
• Title: Personalized Privacy Control in LLMs via Attention Head Intervention
• Abstract summary: We introduce textit personalized privacy, which incorporates user-specific disclosure preferences into privacy control.<n>We also present P3Bench(textbfPersonalized textbfPrivacy textbfPreservation textbfBenchmark), a novel benchmark extending contextual privacy policies with personalized disclosure policies.<n>Our method significantly improves adherence to user-specific privacy preferences by reducing cases where the model fails to follow the given policy.
• Score: 21.49017923823084
• License: http://creativecommons.org/licenses/by-sa/4.0/
• Abstract: The rise of agentic AI enables LLMs to access diverse user data, raising critical privacy concerns. Prior work on contextual privacy studies whether LLMs regulate information disclosure according to context-dependent norms. However, acceptable disclosure boundaries may vary across users even within the same context. To address this limitation, we introduce \textit{personalized privacy}, which incorporates user-specific disclosure preferences into privacy control. We further present P3Bench~(\textbf{P}ersonalized \textbf{P}rivacy \textbf{P}reservation \textbf{Bench}mark), a novel benchmark extending contextual privacy policies with personalized disclosure policies. Experiments show that prompt-based policies fail to reliably enforce personalized privacy policies, with Qwen2.5-7B and Gemma3-4B showing average policy ignorance ratios of 51.25\% and 74.28\%, respectively. Finally, to address this problem, we propose \textsc{Repair}, a robust inference-time attention head intervention method that adjusts disclosure behavior toward policy-consistent responses. Our method significantly improves adherence to user-specific privacy preferences by reducing cases where the model fails to follow the given policy.

Related papers

• Seeing Without Exposing: Adaptive Privacy Control for Open-World, Context-Hungry MLLMs [61.80513991207956]
  We propose Anchored Privacy Drifting (APD) to tackle privacy challenges in MLLMs.<n>APD drifts privacy-sensitive elements toward semantically equivalent alternatives while anchoring contextual cues to the source image.<n>We show that our method achieves balanced improvements in both privacy sanitization and content retention.
  arXiv  Detail & Related papers  (2026-06-05T11:41:39Z)
• Towards Usable Privacy Management for IoT TAPs: Deriving Privacy Clusters and Preference Profiles [1.2744523252873352]
  IoT Trigger-Action Platforms (TAPs) typically offer coarse-grained permission controls.<n>This paper contributes to usable privacy management for TAPs by deriving privacy clusters and profiles for different types of users.
  arXiv  Detail & Related papers  (2025-11-14T12:08:58Z)
• MAGPIE: A dataset for Multi-AGent contextual PrIvacy Evaluation [54.410825977390274]
  Existing benchmarks to evaluate contextual privacy in LLM-agents primarily assess single-turn, low-complexity tasks.<n>We first present a benchmark - MAGPIE comprising 158 real-life high-stakes scenarios across 15 domains.<n>We then evaluate the current state-of-the-art LLMs on their understanding of contextually private data and their ability to collaborate without violating user privacy.
  arXiv  Detail & Related papers  (2025-06-25T18:04:25Z)
• Enhancing Feature-Specific Data Protection via Bayesian Coordinate Differential Privacy [55.357715095623554]
  Local Differential Privacy (LDP) offers strong privacy guarantees without requiring users to trust external parties. We propose a Bayesian framework, Bayesian Coordinate Differential Privacy (BCDP), that enables feature-specific privacy quantification.
  arXiv  Detail & Related papers  (2024-10-24T03:39:55Z)
• PrivacyLens: Evaluating Privacy Norm Awareness of Language Models in Action [54.11479432110771]
  PrivacyLens is a novel framework designed to extend privacy-sensitive seeds into expressive vignettes and further into agent trajectories.<n>We instantiate PrivacyLens with a collection of privacy norms grounded in privacy literature and crowdsourced seeds.<n>State-of-the-art LMs, like GPT-4 and Llama-3-70B, leak sensitive information in 25.68% and 38.69% of cases, even when prompted with privacy-enhancing instructions.
  arXiv  Detail & Related papers  (2024-08-29T17:58:38Z)
• The Privacy Policy Permission Model: A Unified View of Privacy Policies [0.5371337604556311]
  A privacy policy is a set of statements that specifies how an organization gathers, uses, discloses, and maintains a client's data. Most privacy policies lack a clear, complete explanation of how data providers' information is used. We propose a modeling methodology, called the Privacy Policy Permission Model (PPPM), that provides a uniform, easy-to-understand representation of privacy policies.
  arXiv  Detail & Related papers  (2024-03-26T06:12:38Z)
• SeePrivacy: Automated Contextual Privacy Policy Generation for Mobile Applications [21.186902172367173]
  SeePrivacy is designed to automatically generate contextual privacy policies for mobile apps. Our method synergistically combines mobile GUI understanding and privacy policy document analysis. 96% of the retrieved policy segments can be correctly matched with their contexts.
  arXiv  Detail & Related papers  (2023-07-04T12:52:45Z)
• Exploring Consequences of Privacy Policies with Narrative Generation via Answer Set Programming [0.0]
  We present a framework that uses Answer Set Programming (ASP) to formalize privacy policies. ASP allows end-users to forward-simulate possible consequences of the policy in terms of actors. We demonstrate through the example of the Health Insurance Portability and Accountability Act how to use the system in various ways.
  arXiv  Detail & Related papers  (2022-12-13T16:44:46Z)
• Private Reinforcement Learning with PAC and Regret Guarantees [69.4202374491817]
  We design privacy preserving exploration policies for episodic reinforcement learning (RL) We first provide a meaningful privacy formulation using the notion of joint differential privacy (JDP) We then develop a private optimism-based learning algorithm that simultaneously achieves strong PAC and regret bounds, and enjoys a JDP guarantee.
  arXiv  Detail & Related papers  (2020-09-18T20:18:35Z)
• PGLP: Customizable and Rigorous Location Privacy through Policy Graph [68.3736286350014]
  We propose a new location privacy notion called PGLP, which provides a rich interface to release private locations with customizable and rigorous privacy guarantee. Specifically, we formalize a user's location privacy requirements using a textitlocation policy graph, which is expressive and customizable. Third, we design a private location trace release framework that pipelines the detection of location exposure, policy graph repair, and private trajectory release with customizable and rigorous location privacy.
  arXiv  Detail & Related papers  (2020-05-04T04:25:59Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.