LucidiTEE: Scalable Policy-Based Multiparty Computation with Fairness.

Motivated by recent advances in exploring the power of hybridized TEE-blockchain systems, we present LucidiTEE , a unified framework for confidential, policy-compliant computing that guarantees fair output delivery. For context: Ekiden (EuroS &P’19) and FastKitten (Sec’19) use enclave-ledger interactions to enable privacy-preserving smart contracts. However, they store the contract’s inputs on-chain, and therefore, are impractical for applications that process large volumes of data or serve large number of users. In contrast, LucidiTEE implements privacy-preserving computation while storing inputs, outputs, and state off-chain, using the ledger only to enforce policies on computation. Chaudhuri et al. (CCS’17) showed that enclave-ledger interactions enable fair secure multiparty computation. In a setting with n processors each of which possesses a TEE, they show how to realize fair secure computation tolerating up to t corrupt parties for any t < n . We improve upon their result by showing a novel protocol which requires only t out of the n processors to possess a TEE. Kaptchuk et al. (NDSS’19) showed that enclave-ledger interactions can enable applications such as one-time programs and rate limited logging. We generalize their ideas to enforcing arbitrary history-based policies within and across several multi-step computations, and formally specify a new functionality for policy-compliant multiparty computation. Summarizing, LucidiTEE enables multiple parties to jointly compute on private data, while enforcing history-based policies even when input providers are offline, and fairness to all output recipients, in a malicious setting. LucidiTEE uses the ledger only to enforce policies; i.e., it does not store inputs, outputs, or state on the ledger, letting it scale to big data computation. We show novel applications including a personal finance app, collaborative machine learning, and policy-based surveys amongst an apriori-unknown set of participants.