Confidential Data Anonymisation Using Trusted Execution Environments

Data anonymisation is the process of transforming personal or sensitive information within a dataset to prevent the (re-)identification of individuals. By anonymising data, researchers and analysts can make use of statistical patterns in the data without compromising individual privacy. However, when handling large volumes of data, outsourcing the anonymisation process itself defeats the purpose of protecting sensitive information. To address this challenge, building from our existing work on confidential data cleaning [1], we propose the anonymisation of data, containing confidential information, inside a verifiable but outsourced trusted execution environment (TEE). The implementation of the proposed Confidential Data Anonymisation (CDA) system has a web interface, which accepts raw data securely–over secure Hypertext Transfer Protocol (HTTPS)–and transfers it into a remote, verified and attested trusted execution environment. Thereafter, the system performs the anonymisation in the secure environment, and allows the retrieval of the results securely over HTTPS. The implementation uses Intel Software Guard Extensions (SGX) as the confidential computing environment. The data anonymisation functionality is implemented using the Python programming language with no TEE-specific modifications. The code is deployed atop the Gramine library operating system (libOS) and containerised with Docker using the Gramine Shielded Containers (GSC). We demonstrate the practical feasibility of our proposal by presenting a performance evaluation of the work on two Intel SGX-enabled processors.