yak: WebAssembly-based Data Analysis Platform with Data Providers Driven Analysis Control

The increasing reliance on large-scale data analysis raises critical concerns regarding protecting sensitive personal information and regulations like GDPR. While anonymization techniques and contractual agreements offer some safeguards, they do not provide data providers with control over datasets and often hinder data’s flexible and advanced use. This paper introduces "Analysis Control" and its implementation in the "yak" platform, designed to address these challenges. yak enables data providers to define and enforce policies that govern data provisioning and analysis processes, ensuring data confidentiality and controlled utilization. By employing a modular architecture based on WebAssembly and operating within a Trusted Execution Environment, yak allows for secure and extensible analysis. Data providers can verify the platform’s integrity through attestation protocols and control analysis workflows through policies that control analysis function executions and data anonymization. Evaluations using synthetic datasets demonstrate yak’s feasibility, with negligible overhead, and confirm its ability to enforce data provider policies.