Privacy-Preserving Intent-Driven Computing: Minimizing Personal Data Exposure During Autonomous Internet Capability Invocation

The increasing development of autonomous Internet interaction systems, generative artificial intelligence, and heterogeneous digital capability orchestration creates new opportunities for completing complex user tasks through coordinated interactions with multiple websites, APIs, AI models, and third-party services. However, these capabilities also introduce an important privacy challenge: autonomous systems may need to process personal information to accomplish a task, but individual services do not necessarily require access to the same information or to the user's complete personal context. When personal information is transferred indiscriminately across multiple services, unnecessary disclosure may occur even when each individual service receives information that appears relevant to the overall task. This paper proposes Privacy-Preserving Intent-Driven Computing (PPIDC), a conceptual framework designed to minimize unnecessary personal data exposure during autonomous Internet capability invocation. Rather than treating privacy protection exclusively as a static property of applications, websites, or databases, PPIDC treats privacy requirements as dynamic constraints associated with user intentions, task decomposition, individual execution stages, and the information requirements of specific Internet capabilities. The framework introduces Intent-Based Data Minimization (IDDM), Selective Context Disclosure (SCD), Capability-Specific Privacy Policies (CSPP), Sensitive Attribute Filtering (SAF), Privacy-Aware Execution Planning (PAEP), and Cross-Service Information Exposure Measurement (CSIEM). These mechanisms collectively determine which information is necessary for a particular execution step, which information can remain local, which attributes can be transformed into less revealing representations, and what information may be disclosed to an external capability under specified conditions. A central contribution of the proposed framework is the introduction of a privacy-aware capability invocation model in which task execution and information disclosure are jointly optimized. Instead of automatically transmitting the complete contextual information available to an intelligent system, the proposed approach seeks to identify the minimum sufficient information required by each capability while maintaining task feasibility, acceptable execution quality, and user authorization. The framework further introduces an information exposure model for examining cumulative disclosure across multiple services, recognizing that individually limited disclosures may collectively reveal substantial information about a user. Importantly, PPIDC is not intended to replace established privacy technologies such as encryption, anonymization, differential privacy, federated learning, or access control. Instead, it proposes an intent-centered orchestration layer through which these technologies and privacy policies can be incorporated into autonomous Internet execution. This paper develops the framework's conceptual architecture, information disclosure model, privacy-aware decision mechanisms, potential advantages, research propositions, evaluation methodology, and implementation considerations. It presents a conceptual research framework and does not claim completed experimental validation or demonstrated performance superiority.