Personalized Privacy Control in LLMs via Attention Head Intervention

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 personalized privacy, which incorporates user-specific disclosure preferences into privacy control. We further present P3Bench (Personalized Privacy Preservation Benchmark), a benchmark extending contextual privacy policies with personalized disclosure constraints. Experiments show that prompt-based policies fail to reliably enforce personalized privacy constraints, with Qwen2.5-7B and Gemma3-4B showing average policy ignorance ratios of 51.25% and 74.28%, respectively. To address this problem, we propose REPAIR, a novel inference-time attention head intervention method that adjusts disclosure behavior toward policy-consistent responses. Our method significantly reduces both over-refusal and over-sharing, improving adherence to user-specific privacy preferences.